CUDA: quantized GEMM for for IQ2_KS, IQ2_K, IQ3_K #418
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
|
Wow the IQ2_KS improved around 35%!? The 32 block I'd probably try out the larger IQ3_KS and especially IQ5_KS for some mixes in the future if you decide to add them. |
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Outdated suggestions cannot be applied.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.
This PR is a follow up of #417 and (almost) completes the quantized matrix multiplication (a.k.a. MMQ) implementation for
IQX_Kquants. The only one missing isIQ4_KSS, but I don't think I'll do that one as the packing is much too complicated.There are larger performance gains for
IQ2_KS(~35%) than forIQ2_KandIQ3_K(~10%). This is due toIQ2_KShaving blocks of 32 and thus being able to use the more efficient GEMM kernel (see discussion in #417).The graph illustrates the performance improvements for the same setup as in #417.
Looking at this graph and in the graph in #417, I almost feel like adding
IQ3_KSandIQ5_KSas 3- and 5-bit quants with blocks of 32.