Open
Conversation
Condition the trust region mask on advantage sign instead of masking both directions unconditionally. Rename ipo_mask -> dppo_mask in config and internal variables. Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>
mikasenghaas
changed the title
samsja
approved these changes
Apr 2, 2026
![cursor[bot]](https://avatars.githubusercontent.com/in/1210556?s=60&v=4){kind=small}
There was a problem hiding this comment.
Cursor Bugbot has reviewed your changes and found 1 potential issue.
Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.
| ipo_mask_low: Annotated[float, Field(ge=0, description="The low threshold for masking tokens.")] = 0.2 | ||
| ipo_mask_high: Annotated[float, Field(ge=0, description="The high threshold for masking tokens.")] = 0.2 | ||
| dppo_mask_low: Annotated[float, Field(ge=0, description="The low threshold for masking tokens.")] = 0.2 | ||
| dppo_mask_high: Annotated[float, Field(ge=0, description="The high threshold for masking tokens.")] = 0.2 |
There was a problem hiding this comment.
CHANGELOG not updated for breaking config rename
Low Severity
The ipo_mask_low and ipo_mask_high fields in DefaultLossConfig were renamed to dppo_mask_low and dppo_mask_high — a breaking config change — but CHANGELOG.md was not updated. The existing entry at line 127 still references the old ipo_mask_* names. This violates the rule requiring any PR that modifies configuration structures in src/prime_rl/*/config.py to update the changelog.
Triggered by project rule: BugBot Instructions
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
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.
Summary
ipo_mask_{low,high}→dppo_mask_{low,high}in config and internals -- this is BREAKING🤖 Generated with Claude Code
Note
Medium Risk
Changes the core RL objective by altering the default trust-region masking behavior and renaming config fields, which can materially impact training dynamics and is a breaking config change.
Overview
Updates the default RL loss to a DPPO+KL-style objective by conditioning the trust-region token mask on the sign of the advantage instead of masking both probability-increase and probability-decrease violations unconditionally.
Renames loss config thresholds from
ipo_mask_{low,high}todppo_mask_{low,high}(breaking for existing configs) and updates unit tests to use the new fields.Written by Cursor Bugbot for commit 8595ae4. This will update automatically on new commits. Configure here.