ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-02-23 22:54:10 +00:00
Code Issues Packages Projects Releases Wiki Activity

sanitize imatrix: q6_0_r4

Browse Source
This commit is contained in:
Iwan Kawrakow
2025-08-27 14:55:26 +03:00
parent c5de02cffa
commit deb55fff63
1 changed files with 5 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
ggml/src/iqk/iqk_quantize.cpp
View File

@@ -5152,8 +5152,12 @@ size_t quantize_q6_0_r4(const float * src, void * dst, int64_t nrows, int64_t n_
auto row_size_0 = ggml_row_size(GGML_TYPE_Q6_0, n_per_row);
std::vector<char> qtmp(4*row_size_0);
char * qrow = (char *)dst;
QHelper helper(imatrix, n_per_row, 32);
auto q_func = [] (const float * x, void * vy, int n_per_row, const float * imatrix) {
quantize_q6_0(x, (char *)vy, 1, n_per_row, imatrix);
};
for (int row = 0; row < nrows; row += 4) {
quantize_q6_0(src, qtmp.data(), 4, n_per_row, imatrix);
helper.quantize(4, src, qtmp.data(), row_size_0, q_func);
repack_q6_0(4, n_per_row, (const block_q6_0 *)qtmp.data(), (block_q6_0_r4 *)qrow, false);
src += 4*n_per_row;
qrow += 4*row_size_0;