fix: guard zero rollout temperature logprob scaling#2197
Open
EazyReal wants to merge 1 commit into
Open
Conversation
Greedy rollout can use rollout_temperature=0, but both training-side logprob paths divided logits whenever the temperature differed from 1. This produced infinite logits and NaN gradients. Skip temperature scaling when the value is zero while preserving the existing behavior for positive temperatures. Add CPU regression coverage for both call sites and register it in the PR test matrix.
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Both rollout logprob paths divide returned logprobs by the configured sampling temperature. Temperature zero is valid for greedy decoding, but that division produces infinities or NaNs. Skip the scaling only when temperature is zero; nonzero behavior is unchanged. Focused CPU tests cover both call sites.