happyz
/
llama.cpp
mirror of https://github.com/ggerganov/llama.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

ggml : guard against sumq2 being 0 in IQ4_NL (#20460)

This commit is contained in:
Bartowski 2026-03-15 04:47:28 -04:00 committed by GitHub
parent 617db241aa
commit b9da4444df
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 1 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
ggml/src/ggml-quants.c
View File

@ -4767,7 +4767,7 @@ static void quantize_row_iq4_nl_impl(const int super_block_size, const int block
sumqx += w*q*xb[j]; sumqx += w*q*xb[j];
sumq2 += w*q*q; sumq2 += w*q*q;
} }
d = sumqx/sumq2; d = sumq2 > 0 ? sumqx/sumq2 : 0.f;
float best = d*sumqx; float best = d*sumqx;
for (int itry = -ntry; itry <= ntry; ++itry) { for (int itry = -ntry; itry <= ntry; ++itry) {
id = (itry + values[0])/max; id = (itry + values[0])/max;