vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Guard against all weights in a super-block being zero

Browse source
This commit is contained in:
Iwan Kawrakow 2023-09-04 14:10:15 +03:00
parent 3103568144
commit 79f1757661
1 changed files with 6 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

6
k_quants.c
View file

@ -1086,6 +1086,12 @@ void quantize_row_q6_K_reference(const float * restrict x, block_q6_K * restrict
} }
if (!max_abs_scale) {
memset(&y[i], 0, sizeof(block_q6_K));
y[i].d = ggml_fp32_to_fp16(0.f);
continue;
}
float iscale = -128.f/max_scale; float iscale = -128.f/max_scale;
y[i].d = ggml_fp32_to_fp16(1/iscale); y[i].d = ggml_fp32_to_fp16(1/iscale);
for (int ib = 0; ib < QK_K/16; ++ib) { for (int ib = 0; ib < QK_K/16; ++ib) {