diff options
| author | Iwan Kawrakow <iwan.kawrakow@gmail.com> | 2024-07-30 16:11:25 +0300 |
|---|---|---|
| committer | Kawrakow <48489457+ikawrakow@users.noreply.github.com> | 2024-08-01 09:38:06 +0200 |
| commit | 4f237d44f6d75afbb5cef39d4d6b0b35b2a517c7 (patch) | |
| tree | 097eb7ee5be6ec08d3958f747775f9225db3c7a4 /include/llama.h | |
| parent | 36204c4ec7815f1b0f8f38c8e1cd3e6b3fef2eaa (diff) | |
iq3_k: Basics
Quantize/dequantize, CUDA dequantize.
PPL of LLaMA-3.1-8B is better than iq3_s and iq3_m.
Diffstat (limited to 'include/llama.h')
| -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 7bccd4bb..88d82958 100644 --- a/include/llama.h +++ b/include/llama.h @@ -171,8 +171,9 @@ extern "C" { LLAMA_FTYPE_MOSTLY_IQ1_BN = 36, // except 1d tensors LLAMA_FTYPE_MOSTLY_IQ2_BN = 37, // except 1d tensors LLAMA_FTYPE_MOSTLY_IQ2_K = 38, // except 1d tensors - LLAMA_FTYPE_MOSTLY_IQ4_K = 39, // except 1d tensors - LLAMA_FTYPE_MOSTLY_IQ5_K = 40, // except 1d tensors + LLAMA_FTYPE_MOSTLY_IQ3_K = 39, // except 1d tensors + LLAMA_FTYPE_MOSTLY_IQ4_K = 40, // except 1d tensors + LLAMA_FTYPE_MOSTLY_IQ5_K = 41, // except 1d tensors LLAMA_FTYPE_GUESSED = 1024, // not specified in the model file }; |