From 4b53bc876eda6d89890711998b3ab5bf033b2199 Mon Sep 17 00:00:00 2001 From: Kawrakow Date: Fri, 20 Dec 2024 12:02:42 +0100 Subject: IQ2_XXS_R4 (#154) * iq2_xxs_r4: Zen4 Disapointing gain: 134.7 t/s -> 151.1 t/s for PP-512 TG-128 is better: 3.45 -> 4.61 t/s @ 1 thread * Minor * iq2_xxs_r4: NEON --------- Co-authored-by: Iwan Kawrakow --- examples/quantize/quantize.cpp | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) (limited to 'examples') diff --git a/examples/quantize/quantize.cpp b/examples/quantize/quantize.cpp index b69b8a6b..1ad5108e 100644 --- a/examples/quantize/quantize.cpp +++ b/examples/quantize/quantize.cpp @@ -22,6 +22,7 @@ static const std::vector QUANT_OPTIONS = { { "Q5_1", LLAMA_FTYPE_MOSTLY_Q5_1, " 4.70G, +0.0349 ppl @ LLaMA-v1-7B", }, { "Q6_0", LLAMA_FTYPE_MOSTLY_Q6_0, " 6.5 bpw quantization", }, { "IQ2_XXS", LLAMA_FTYPE_MOSTLY_IQ2_XXS, " 2.06 bpw quantization", }, + { "IQ2_XXS_R4",LLAMA_FTYPE_MOSTLY_IQ2_XXS_R4,"IQ2_XXS repacked", }, { "IQ2_XS", LLAMA_FTYPE_MOSTLY_IQ2_XS, " 2.31 bpw quantization", }, { "IQ2_S", LLAMA_FTYPE_MOSTLY_IQ2_S, " 2.5 bpw quantization", }, { "IQ2_M", LLAMA_FTYPE_MOSTLY_IQ2_M, " 2.7 bpw quantization", }, @@ -503,7 +504,7 @@ int main(int argc, char ** argv) { if (!params.ignore_imatrix_rules && imatrix_data.empty() && (params.ftype == LLAMA_FTYPE_MOSTLY_IQ2_XS || params.ftype == LLAMA_FTYPE_MOSTLY_IQ2_XXS || - params.ftype == LLAMA_FTYPE_MOSTLY_IQ2_S || + params.ftype == LLAMA_FTYPE_MOSTLY_IQ2_S || params.ftype == LLAMA_FTYPE_MOSTLY_IQ2_XXS_R4 || params.ftype == LLAMA_FTYPE_MOSTLY_Q2_K_S || params.ftype == LLAMA_FTYPE_MOSTLY_IQ1_S || params.ftype == LLAMA_FTYPE_MOSTLY_IQ1_M)) { -- cgit v1.2.3