diff options
Diffstat (limited to 'examples')
| -rw-r--r-- | examples/quantize/quantize.cpp | 3 |
1 files changed, 2 insertions, 1 deletions
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<struct quant_option> 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)) { |