Fix KV cache handling for Gemma4 26B - #1235
Draft
nsivalin wants to merge 2 commits into
Draft
Conversation
nsivalin
marked this pull request as ready for review
August 5, 2026 03:55
Contributor
|
Please keep this PR in draft unless a maintainer approves it. Then you can make it ready for review. |
ochougul
marked this pull request as draft
August 5, 2026 04:38
Contributor
|
Can you fix the LINT issues |
quic-rishinr
force-pushed
the
fix/gemma4-26b-kv-fix
branch
from
August 7, 2026 04:43
814192a to
1bbcf0f
Compare
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
3 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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Fixes sliding-window KV cache corruption in
QEffGemma4DynamicLayer.update()where physical-slot validity was computed from the wrapped/physical position instead of the logical position, causing attention to read stale/garbage KV data (leading to infinite repetition and non-deterministic output) once a request's sequence length crossed thesliding_windowboundary.The fix is scoped to
google/gemma-4-26B-A4B-itonly (viaconfig.num_expertsas the differentiator, leaving the original code path unchanged for all other Gemma4 variants (31B/E2B/E4B).