llama.cpp

d15d177f - vulkan: faster q6_k matmul (#17813)

vulkan: faster q6_k matmul (#17813) * q6_k faster mul mat * 8 values * fix comment * switch to two at a time * start ci for .glsl files