fix: reject non-positive rollout temperature at parse time - #2297
Merged
zhuzilin merged 1 commit intoAug 21, 2026
Conversation
Greedy T=0 is not a usable on-policy RL distribution. Fail in slime_validate_args instead of special-casing it later in logprob math. Eval can still use --eval-temperature 0. Co-authored-by: Cursor <cursoragent@cursor.com>
2 tasks
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Problem
--rollout-temperature 0is greedy decoding. That is not a usable on-policy RL policy: the sampling distribution is a delta, while log-probs (and PPO ratios) still come from a softmax. Greedy RL should be rejected at config parse, not assumed away in loss code.--eval-temperature 0is unchanged; greedy eval remains valid.Change
slime_validate_argsraises ifrollout_temperature <= 0.Test plan
tests/test_megatron_argument_validation.py—0.0and-0.1raise; existing validate-args cases still pass at the default1.0Made with Cursor