Implement sliding attention in Gemma3 #11409
Merged
Conversation
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
woct0rdho
requested review from
Kosinkadink,
comfyanonymous and
guill
as code owners
lrivera
pushed a commit
to Research-Warrant/ComfyUI
that referenced
this pull request
Jan 8, 2026
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Outdated suggestions cannot be applied.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.
As the NewBie model uses detailed XML prompts that can be > 1024 tokens, it's time to implement the sliding attention that's previously missing in Gemma3.
I found a mistake in the config that Gemma3 should use 5 sliding + 1 non-sliding, not vice-versa.
rope_thetaandrope_scaleare also swapped accordingly. (Before sliding attention is implemented, this does not affect the results with < 1024 tokens.)After this PR, the condition tensor from Gemma3 in ComfyUI is much closer to the one from Gemma3 in Transformers.