diff options
| author | Kawrakow <iwankawrakow@gmail.com> | 2024-12-11 18:16:49 +0100 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2024-12-11 18:16:49 +0100 |
| commit | 8c6b84220d78d1742cbe55e21746ad68c8df267a (patch) | |
| tree | f6015f704e2dca4633a9fafa82c80508c7256e99 /include | |
| parent | 9469af87f71a8d174064e142ae8eff390d387c6c (diff) | |
Q2_K_R4 (#136)
* q2_k_r4: Zen4
PP-512(LLaMA-3.1-8B) = 256 t/s
* q3_k_r4: AVX2
* q2_k_r4: AVX2
We get PP-512(LLaMA-3.1-8B) = 287 t/s.
Also cherry-picked the q3_k_r4 AVX2 adaptation that I somehow
forgot to push upstream.
* q2_k_r4: NEON
We get PP-512(LLaMA-3.1-8B) = 106.2 t/s.
TG-128 is 36.02 t/s, which is ~10% higher than q2_K_S.
* Make sure rows per thread are a multiple of 4
---------
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
Diffstat (limited to 'include')
| -rw-r--r-- | include/llama.h | 1 |
1 files changed, 1 insertions, 0 deletions
diff --git a/include/llama.h b/include/llama.h index f87d13ff..0992b10a 100644 --- a/include/llama.h +++ b/include/llama.h @@ -183,6 +183,7 @@ extern "C" { LLAMA_FTYPE_MOSTLY_Q4_0_R4 = 202, // except 1d tensors LLAMA_FTYPE_MOSTLY_Q8_0_R4 = 207, // except 1d tensors LLAMA_FTYPE_MOSTLY_Q5_0_R4 = 208, // except 1d tensors + LLAMA_FTYPE_MOSTLY_Q2_K_R4 = 210, // except 1d tensors LLAMA_FTYPE_MOSTLY_Q3_K_R4 = 211, // except 1d tensors LLAMA_FTYPE_MOSTLY_Q4_K_R4 = 214, // except 1d tensors LLAMA_FTYPE_MOSTLY_Q5_K_R4 = 216, // except 1d tensors |