llama.cpp

d6d0ce82 - vulkan: reduce iq1 shared memory usage for mul_mm (#24287)