diff options
| author | Kawrakow <iwankawrakow@gmail.com> | 2025-02-19 11:47:07 +0200 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2025-02-19 11:47:07 +0200 |
| commit | a0ebfdd661a2ccb2700b0e36cfc10ca1a2b4de98 (patch) | |
| tree | d5bb2c8f07625c617d1113348b4b67d79b8f64f4 /include/llama.h | |
| parent | 047ba895bb3d94f055756c1ec7767b3342cb9c90 (diff) | |
Q8_KV: 8-bit quantization type targeting the KV cache (#208)
* Adding q8_KV - Basics + AVX2 gemm/gemv
* q8_KV: Better AVX2 gemm
* q8_KV: Better Zen4 gemm
We get 225.7 t/s for L3-8B. In comparison q8_0 without
run-tinme-repacking is at 169 t/s.
* q8_KV: AVX2 gemm/gemv
We get 254 t/s for L3-8B vs 194 t/s for q8_0 without rtr.
* q8_KV: be able to use it for K cache
This required quite a few fixes in ggml and llama.cpp:
* ggml: do not calculate row size as n/block_size*type_size. I had
removed most of it when implementing the quants with per row scale,
bit it was stull lurking in ggml_copy. Not sure if these were the last
remnants of ggmil-style row sizes, or if there are still places left
* llama.cpp: get rid of the the 1d K cache assumption. Create and manage
the K-cache as a 2D tensor so we can have per row meta data as needed
by q8_KV.
Using q8_KV for K-cache results in non-negligible performance gains.
More details to follow, but for DeepSeek-Lite with MLA, we get
18% speedup for PP-8192 compared to q8_0 K-cache.
* q8_KV: be able to use it for K cache in FA
* q8_KV: repack it for K*Q in FA
* q8_KV: slightly faster gemv on Zen4
* q8_KV: slightly faster gemv on Zen4
* q8_KV: ARM_NEON
We get PP-512 = 167 t/s for L3-8B without interleaving!
We do the interleaving on the fly, so I wonder if this
could be done for other quants as well.
* q8_KV: use it in FA on NEON
* q8_KV_r8 - repacked q8_KV
On Zen4 it is slower than q8_k_r8 (292 vs 370 t/s)
This makes no sense whatsoever as the q8_KV_r8 GEMM is
basically the q8_k_r8 GEMM with the unnecessary block stuff
removed (so, one would think that it would be faster).
* q8_KV_r8: don't use nrc_y = 16 on Zen4
This is faster - 350 t/s. Why?
Much better than the 290 t/s we had before, but still slower
than the 370 t/s for q8_k_r8.
* q8_KV: nrc_y = 16 also doesn't pay off in FA
* Minor
---------
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
Diffstat (limited to 'include/llama.h')
| -rw-r--r-- | include/llama.h | 2 |
1 files changed, 2 insertions, 0 deletions
diff --git a/include/llama.h b/include/llama.h index 39251d35..b5ad65e7 100644 --- a/include/llama.h +++ b/include/llama.h @@ -180,6 +180,7 @@ extern "C" { LLAMA_FTYPE_MOSTLY_IQ3_KL = 146, // except 1d tensors LLAMA_FTYPE_MOSTLY_IQ2_KS = 147, // except 1d tensors LLAMA_FTYPE_MOSTLY_IQ4_KSS = 148, // except 1d tensors + LLAMA_FTYPE_MOSTLY_Q8_KV = 149, // except 1d tensors // LLAMA_FTYPE_MOSTLY_Q4_0_R8 = 202, // except 1d tensors LLAMA_FTYPE_MOSTLY_Q8_0_R8 = 207, // except 1d tensors @@ -206,6 +207,7 @@ extern "C" { LLAMA_FTYPE_MOSTLY_IQ4_K_R4 = 340, // except 1d tensors LLAMA_FTYPE_MOSTLY_IQ5_K_R4 = 341, // except 1d tensors LLAMA_FTYPE_MOSTLY_IQ4_KS_R4 = 345, // except 1d tensors + LLAMA_FTYPE_MOSTLY_Q8_KV_R8 = 398, // except 1d tensors LLAMA_FTYPE_MOSTLY_Q8_K_R8 = 399, // except 1d tensors LLAMA_FTYPE_GUESSED = 1024, // not specified in the model file |