From a867b919ca1e26cc828f98c35b4c6926e8e54762 Mon Sep 17 00:00:00 2001 From: Kawrakow Date: Sat, 21 Dec 2024 08:32:39 +0100 Subject: IQ2_XS_R4 (#155) * iq2_xs_r4: Zen4 * iq2_xs_r4: AVX2 * iq2_xs_r4: slightly better matrix x vector on AVX2 * iq2_xs_r4: NEON - not much better than iq2_xs * iq2_xs_r4: slightly better 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 1ad5108e..dbae9792 100644 --- a/examples/quantize/quantize.cpp +++ b/examples/quantize/quantize.cpp @@ -24,6 +24,7 @@ static const std::vector QUANT_OPTIONS = { { "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_XS_R4",LLAMA_FTYPE_MOSTLY_IQ2_XS_R4,"IQ2_XS repacked", }, { "IQ2_S", LLAMA_FTYPE_MOSTLY_IQ2_S, " 2.5 bpw quantization", }, { "IQ2_M", LLAMA_FTYPE_MOSTLY_IQ2_M, " 2.7 bpw quantization", }, { "IQ1_S", LLAMA_FTYPE_MOSTLY_IQ1_S, " 1.56 bpw quantization", }, @@ -505,7 +506,7 @@ 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_XXS_R4 || - params.ftype == LLAMA_FTYPE_MOSTLY_Q2_K_S || + params.ftype == LLAMA_FTYPE_MOSTLY_Q2_K_S || params.ftype == LLAMA_FTYPE_MOSTLY_IQ2_XS_R4 || params.ftype == LLAMA_FTYPE_MOSTLY_IQ1_S || params.ftype == LLAMA_FTYPE_MOSTLY_IQ1_M)) { fprintf(stderr, "\n==========================================================================================================\n"); -- cgit v1.2.3