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

iq3_xs: minor fix

Browse source
This commit is contained in:
Iwan Kawrakow 2024-02-27 08:52:49 +02:00
parent ad40ae635f
commit 6c2b233b08
1 changed files with 1 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

2
ggml-quants.c
View file

@ -12240,7 +12240,7 @@ static void quantize_row_iq4_nl_impl(const int super_block_size, const int block
if (super_block_size/block_size > 1) { if (super_block_size/block_size > 1) {
int nb = super_block_size/block_size; int nb = super_block_size/block_size;
memset(scales_h, 0, ((nb+3)/4)*sizeof(uint16_t)); memset(scales_h, 0, ((nb+7)/8)*sizeof(uint16_t));
float d = -max_scale/32; float d = -max_scale/32;
dh[0] = GGML_FP32_TO_FP16(d); dh[0] = GGML_FP32_TO_FP16(d);
float id = d ? 1/d : 0.f; float id = d ? 1/d : 0.f;