ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-01-26 17:20:01 +00:00
Code Issues Packages Projects Releases Wiki Activity

Fix #772 (#790)

Browse Source 
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
This commit is contained in:
Kawrakow
2025-09-23 16:43:02 +02:00
committed by GitHub
parent 079231c291
commit 8b4208e789
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
ggml/src/ggml-cuda/fattn-mma-f16.cuh
View File

@@ -1408,7 +1408,7 @@ void launch_fattn_mma(
//const bool use_stream_k = cc >= CC_ADA_LOVELACE || tiles_efficiency_percent < 75;
// On my RTX-4080 the above is slightly slower for PP. It would be useful to try and see what happens on Blackwell
const bool use_stream_k = tiles_efficiency_percent < 75;
const bool use_stream_k = tiles_efficiency_percent < 75 || Q->ne[1] > 2048;
blocks_num.x = use_stream_k ? nblocks_stream_k : ntiles_total;
blocks_num.y = 1;