From 4f237d44f6d75afbb5cef39d4d6b0b35b2a517c7 Mon Sep 17 00:00:00 2001 From: Iwan Kawrakow Date: Tue, 30 Jul 2024 16:11:25 +0300 Subject: iq3_k: Basics Quantize/dequantize, CUDA dequantize. PPL of LLaMA-3.1-8B is better than iq3_s and iq3_m. --- include/llama.h | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) (limited to 'include/llama.h') 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 }; -- cgit v1.2.3