Mirrors/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2024-12-25 22:08:46 +01:00
Code Issues Packages Projects Releases Wiki Activity

ggml : fix another case of quants nans (#7387)

Browse Source
This commit is contained in:
slaren 2024-05-19 17:08:46 +02:00 committed by GitHub
parent 5ca49cbecd
commit e4e6f67be6
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-quants.c
View File

@ -1149,7 +1149,7 @@ static float make_qx_quants(int n, int nmax, const float * restrict x, int8_t *
sumlx += w*x[i]*l; sumlx += w*x[i]*l;
suml2 += w*l*l; suml2 += w*l*l;
} }
float scale = sumlx/suml2; float scale = suml2 ? sumlx/suml2 : 0.0f;
if (return_early) return suml2 > 0 ? 0.5f*(scale + 1/iscale) : 1/iscale; if (return_early) return suml2 > 0 ? 0.5f*(scale + 1/iscale) : 1/iscale;
float best = scale * sumlx; float best = scale * sumlx;
for (int is = -9; is <= 9; ++is) { for (int is = -9; is <= 9; ++is) {