llama.cpp

vulkan: scale caching for k quants + misc fixes

#11081

Merged