llama.cpp

4064c0e3 - Vulkan: Use improved q4_k and q5_k dequant code in dequant shaders (#10798)