Fix exact-zero DiffusionGemma self-conditioning - #776
Closed
sylvesterkaczmarek wants to merge 1 commit into
Closed
Conversation
Address the self-conditioning semantics reported in google-deepmind#773. Zero vocabulary logits become a nonzero vocabulary-mean embedding after encode_logits, so they cannot represent the report-defined disabled self-conditioning signal. Preserve the existing all-zero-logit sentinel across the embedding boundary and make the no-previous-logits first pass exactly zero as well. Add regression coverage for the absent signal, the mixed enabled/disabled batch case, and an explicit enable mask.
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Address the self-conditioning semantics reported in #773.
Zero vocabulary logits become a nonzero vocabulary-mean embedding after encode_logits, so they cannot represent the report-defined disabled self-conditioning signal. Preserve the existing all-zero-logit sentinel across the embedding boundary and make the no-previous-logits first pass exactly zero as well.
Add regression coverage for the absent signal, the mixed enabled/disabled batch case, and an explicit enable mask.