ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-04-30 19:31:48 +00:00
Code Issues Packages Projects Releases Wiki Activity

Minor

Browse Source
This commit is contained in:
Iwan Kawrakow
2025-08-27 16:57:54 +03:00
parent 756f3df7d3
commit c9b50fd45c
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
ggml/src/iqk/iqk_quantize.cpp
View File

@@ -7609,7 +7609,7 @@ size_t quantize_iq1_s_r4(const float * src, void * dst, int64_t nrows, int64_t n
weight[j] = imatrix[kBlockSize*ibl + j]*sqrt(sigma2 + xb[j]*xb[j]); weight[j] = imatrix[kBlockSize*ibl + j]*sqrt(sigma2 + xb[j]*xb[j]);
sumwx += weight[j]*std::abs(xb[j]); sumwx += weight[j]*std::abs(xb[j]);
} }
if (!sumwx) { if (sumwx < 1e-14f) {
printf("Found block with mismatching importance/model weights\n"); printf("Found block with mismatching importance/model weights\n");
// Either all weights are zero, or xb is zero where weight is not zero. // Either all weights are zero, or xb is zero where weight is not zero.
// In both of these cases it is better to simply ignore the imatrix // In both of these cases it is better to simply ignore the imatrix