llama.cpp
vulkan: Support FA with any multiple of 8 head sizes
#15537
Merged

Loading