llama.cpp

vulkan: fix coopmat2 flash attention for non-contiguous inputs

#11281

Merged