Q8_K_R8: Fastest quantized matrix multiplications #141
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 5 commits
December 13, 2024 18:21
We get PP-512(LLaMA-3.1-8B) = 370 t/s on a Ryzen-7950X!
I was worried that we don't have enough vector registrers on AVX2, but it looks like it handles it just fine. We get PP-512(LLaMA-3.1-8B) = 354 t/s on a Ryzen-5975WX. Slightly slower than the Zen4 version with double the threads, but still a huge upgrade compared to Q8_0_R4.
We get PP-512(LLaMA-3.1-8B) = 159.2 t/s. Compare this to the 128 t/s we have fr Q8_0_R4.
Why? * On AVX2 _mm256_maddubs_epi16() may overflow, so we need to stay within the signed int range and use _mm256_sign_epi8. Not yet tested on the AVX2 comp, vut expect major slowdown. * It is almost 10% faster on ARM_NEON. Somehow the veorrq_u8() needed tto convert from unsigned to signed seems to be extremely slow on the M2-Max * We only lose ~0.5% in oerformance on Zen4 (there the exclusive or that we now use to convert fro signed to unsigned seems to be much faster than on M2-Max)
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Outdated suggestions cannot be applied.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 adds
Q8_K_R8- 8-rows interleaved version ofQ8_K. With that, we break the world record in prompt processing speed. Here is what we get for PP-512 with LLaMA-3.1-8B onZen4(Ryzen-7950X),AVX2(Ryzen-5975WX) andARM_NEON(M2-Max):On the Ryzen-7950X, which provides native
bf16support, this is nearly 60% faster thanbf16. On the M2-Max, which has nativefp16support,Q8_K_R8is 87% faster thanfp16!Note on AVX2: In the
AVX2implementation one needs to use the_mm256_madd_epi16(x, y)instruction, wherexholds unsigned 8-bit integers andyhas signed 8-bit integers. In the initial implementation I forgot for the 177'th time that the unsigned integers still need to be within0...127, else adding up two adjacent products (as the instruction does) may overflow theint16_trange (and gets silently truncated if it does), so I was making theQ8_K_R8quants unsigned (simplyxor 0x80). This implementation resulted in 354 t/s on the Ryzen-5975WX. Sadly, one needs to "unsign" theQ8_K_R8quants with_mm256_sign_epi8(x, x), and then apply the sign to the activation quants before taking the dot product. This is quite costly andAVX2performance drops to 293 t/s. Being curious about the effect that theint16_toverflow might have, I computed LLaMA-3.1-8B-Instruct perplexity (context 512 tokens) with the original and with the correct implementation. I getPPL = 7.3725with the overflowing variant, andPPL = 7.3443with the correct implementation. I.e., the effect is small but noticeable.