From 2700d3af36e5ce11f6c2c9644130635f83bae6d5 Mon Sep 17 00:00:00 2001 From: Kawrakow Date: Thu, 12 Dec 2024 16:04:20 +0100 Subject: 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 --- include/llama.h | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) (limited to 'include') 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 }; -- cgit v1.2.3