[CI] assert train_rollout_logprob_abs_diff <= 0.1 (regression guard) - #94
Merged
Conversation
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
Code Review
This pull request introduces a CI test assertion in slime/backends/megatron_utils/model.py to ensure that the absolute difference in training rollout log probabilities (train/train_rollout_logprob_abs_diff) is less than or equal to 0.1. There are no review comments, and I have no feedback to provide.
This was referenced May 31, 2026
Backports slime #1968. We already report train/train_rollout_logprob_abs_diff (loss.py) but never assert on it. This adds the CI guard so the train/rollout logprob divergence that #69 fixed cannot silently regress. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Signed-off-by: aoshen02 <aoshen@inferact.ai>
aoshen02
force-pushed
the
sync-slime-logprob-abs-diff-assert
branch
from
May 31, 2026 02:56
658a412 to
765122d
Compare
momo609
pushed a commit
that referenced
this pull request
Jun 8, 2026
…94) Backports slime #1968. We already report train/train_rollout_logprob_abs_diff (loss.py) but never assert on it. This adds the CI guard so the train/rollout logprob divergence that #69 fixed cannot silently regress. Signed-off-by: aoshen02 <aoshen@inferact.ai> Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
What
Backports slime #1968.
We already compute and report
train/train_rollout_logprob_abs_diff(the mean abs diff between training-time and rollout-time log-probs) inloss.py, but nothing asserts on it. This adds the CI guard:Why
vime #69 ("[Bugfix] Fix high train_rollout_logprob_abs_diff") fixed exactly this divergence. Without an assertion, the same class of regression can creep back in silently. This is the missing regression guard for #69 —
--ci-testruns now fail fast if train/rollout logprobs diverge past 0.1.Decoupling
Independent of #92 (GLM bridge) and #93 (R3 KL guards). Single 3-line insert, no overlap with the lines #93 touches in this file.
Test
ast.parseclean. Guard only fires under--ci-testwhen the metric is present.🤖 Generated with Claude Code