ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-04-29 02:41:47 +00:00
Code Issues Packages Projects Releases Wiki Activity

Minor tweaks

Browse Source
This commit is contained in:
Iwan Kawrakow
2024-11-18 15:08:12 +02:00
parent 5705dc7f2e
commit 2be4cffe66
2 changed files with 16 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
ggml/src/iqk/iqk_quantize.cpp
View File

@@ -4036,7 +4036,8 @@ void quantize_row_iq3_kt_impl(const float * x, void * vy, int n_per_row, const f
scales[ib] = 0;
if (!amax) continue;
float scale_0 = std::max(80.f, 127.f*amax/amax_row);
float scale_0 = std::max(80.f, 123.f*amax/amax_row);
//float scale_0 = 80.f;
float best = 0;
for (int itry = -5; itry <= 5; ++itry) {
quantizer.find_best_match(amax/(scale_0 + kStep*itry), xb, weight, best_idx);