ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-01-26 17:20:01 +00:00
Code Issues Packages Projects Releases Wiki Activity

IQ2_XS_R4 (#155)

Browse Source 
* 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 <iwan.kawrakow@gmail.com>
This commit is contained in:
Kawrakow
2024-12-21 08:32:39 +01:00
committed by GitHub
parent 9dfd69bd93
commit 6892554e43
10 changed files with 363 additions and 18 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
examples/quantize/quantize.cpp
View File

@@ -24,6 +24,7 @@ static const std::vector<struct quant_option> 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");