Fix NEON gemm/gemv for legacy quants when row size is not divisible by 128#213
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
I have broken it quite a while ago when I changed the NEON implementation to do two rows at a time. I haven't noticed as all models I typically use have row sizes that are multiple of 128. But as I was working on the
IQ1_SNEON implementation for PR #212, I was testing with DeepSeek-Lite (where K cache row size is 576, so not divisible by 128), usingQ8_0for K cache (but no FA, where it works), and was getting NaNs or gibberish. I lost so much time until I finally realized that the issue is with the K cacheQ8_0matrix multiplication rather than myIQ1_Simplementation.This PR fixes this.