ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-02-19 12:50:11 +00:00
Code Issues Packages Projects Releases Wiki Activity

Do not repack q8_0 for batch sizes less than 8

Browse Source
This commit is contained in:
Kawrakow
2026-02-02 09:07:45 +00:00
parent a527b5af25
commit d5498c4467
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_flash_attn.cpp
View File

@@ -152,7 +152,7 @@ extern "C" IQK_API bool iqk_flash_attn_noalibi(int type_q, int type_mask, float
int int_type_k = int_type_k_in;
auto work_buffer = work_buffer_in;
if (neq1 >= 8 || (rk2 >= 8 && nek2 > 1)) {
if (neq1 >= 8) {
uint64_t row_size = 0;
work_buffer = iqk_repack_k(int_type_k, Dk, nek1, nek2, nek3, stride_k, nbk2, nbk3, k, work_buffer_in, ith, nth, int_type_k, row_size);
if (int_type_k != int_type_k_in) {