NEON implementation for trellis quants #471
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
added 6 commits
May 24, 2025 18:25
Have to use f32 arithmetic else I get gibberish? Correspondigly ridiculously slow.
Closed
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Outdated suggestions cannot be applied.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.
Alternative to #460
One wouldn't really want to use this on a NEON CPU as it is much too slow. But for the sake of completeness, here it is.
Sweep bench results for LLaMA-3.1-8B-Instruct with BLAS on M2-Max CPU (PP performance is much lower without BLAS)
IQ2_KT
IQ3_KT
IQ4_KT
This is nevertheless quite a bit faster than #460, so I'll go with this PR.
Of note: I couldn't make
IQ4_KTwork withfp16arithmetic for some reason. Not sure if there really isfp16range overflow, or if I just have a bug in thefp16implementation that I simply cannot see.