llama.cpp

706e3f93 - vulkan: Implement mmvq for iq1_s/iq1_m (#18450)