From f4170f78bdfe69050c60bf3b51370d6c97593489 Mon Sep 17 00:00:00 2001 From: Kawrakow Date: Fri, 20 Dec 2024 09:12:48 +0100 Subject: IQ3_XXS_R4 (#153) * iq3_xxs_r4: 1st shot on Zen4 PP-512: 107 t/s -> 137 t/s TG-128(1 thread): 2.64 t/s -> 3.44 t/s * iq4_xxs_r4: WIP * iq4_xxs_r4: 1st shot at AVX2 Note: there is a bug in the AVX2 implementation for nrc_y = 1 for IQ quants with blocks of 32. I have fixed it for now by using the nrc_y > 1 implementation (which works) also for nrc_y = 1. * iq3_xxs_r4: NEON --------- Co-authored-by: Iwan Kawrakow --- examples/quantize/quantize.cpp | 1 + 1 file changed, 1 insertion(+) (limited to 'examples/quantize') diff --git a/examples/quantize/quantize.cpp b/examples/quantize/quantize.cpp index cefe2735..b69b8a6b 100644 --- a/examples/quantize/quantize.cpp +++ b/examples/quantize/quantize.cpp @@ -34,6 +34,7 @@ static const std::vector QUANT_OPTIONS = { { "Q2_K_R4", LLAMA_FTYPE_MOSTLY_Q2_K_R4, "Q2_K_S repacked", }, { "Q2_K_S", LLAMA_FTYPE_MOSTLY_Q2_K_S, " 2.16G, +9.0634 ppl @ LLaMA-v1-7B", }, { "IQ3_XXS", LLAMA_FTYPE_MOSTLY_IQ3_XXS, " 3.06 bpw quantization", }, + { "IQ3_XXS_R4",LLAMA_FTYPE_MOSTLY_IQ3_XXS_R4,"IQ3_XXS repacked", }, { "IQ3_S", LLAMA_FTYPE_MOSTLY_IQ3_S, " 3.44 bpw quantization", }, { "IQ3_M", LLAMA_FTYPE_MOSTLY_IQ3_M, " 3.66 bpw quantization mix", }, { "Q3_K", LLAMA_FTYPE_MOSTLY_Q3_K_M, "alias for Q3_K_M" }, -- cgit v1.2.3