Repository navigation
Fix multi-token attention mask in RotatingKVCache - #1978
Open
robertcprice wants to merge 1 commit into
Open
robertcprice wants to merge 1 commit into
robertcprice wants to merge 1 commit into
Conversation
This branch has not been deployed
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
The RotatingKVCache attention mask was incorrectly masking out the keep sink tokens during chunked prefill. By setting the sliding window width to max_size - keep, we ensure that the query attends to exactly max_size tokens: the keep prefix tokens plus the most recent max_size - keep tokens. I've verified the fix with unit tests, confirming that the logit alignment between step-by-step decoding and chunked prefill is now consistent.
I used an AI coding assistant to help isolate the mask overlap bug and write the corresponding unit tests, while I reviewed and verified all logic and results.