diff options
| author | Kawrakow <iwankawrakow@gmail.com> | 2024-12-12 16:04:20 +0100 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2024-12-12 16:04:20 +0100 |
| commit | 2700d3af36e5ce11f6c2c9644130635f83bae6d5 (patch) | |
| tree | 7e3a65fd9c8318fd654f83f7431b906cc017b34a /include | |
| parent | aecc95c0cabc6604642e7bc4a8c9e5cb5233ebc4 (diff) | |
IQ4_K_R4 (#138)
* iq4_k_r4: WIP
* iq4_k_r4: Zen4 and hopefully AVX2
On Zen4 we get PP-512(LLaMA-3.1-8B) = 232.6 t/s, up from 182.2 t/s
for iq4_k. Applying the extra shift costs a ~6 performance penalty.
* iq4_k_r4: AVX2
PP-512 = 227.60 t/s. The shifts are really costly.
* iq4_k_r4: NEON
We get PP-512(LLaMA-3.1-8B) = 108 t/s, up from 58.2 t/s for iq4_k.
---------
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
Diffstat (limited to 'include')
| -rw-r--r-- | include/llama.h | 5 |
1 files changed, 3 insertions, 2 deletions
diff --git a/include/llama.h b/include/llama.h index 0992b10a..117f1659 100644 --- a/include/llama.h +++ b/include/llama.h @@ -190,8 +190,9 @@ extern "C" { LLAMA_FTYPE_MOSTLY_Q6_K_R4 = 218, // except 1d tensors LLAMA_FTYPE_MOSTLY_IQ4_NL_R4 = 225, // except 1d tensors LLAMA_FTYPE_MOSTLY_IQ4_XS_R4 = 230, // except 1d tensors - LLAMA_FTYPE_MOSTLY_Q6_0_R4 = 235, // except 1d tensors - LLAMA_FTYPE_MOSTLY_IQ2_BN_R4 = 237, // except 1d tensors + LLAMA_FTYPE_MOSTLY_Q6_0_R4 = 335, // except 1d tensors + LLAMA_FTYPE_MOSTLY_IQ2_BN_R4 = 337, // except 1d tensors + LLAMA_FTYPE_MOSTLY_IQ4_K_R4 = 340, // except 1d tensors LLAMA_FTYPE_GUESSED = 1024, // not specified in the model file }; |