llama.cpp

70f806b8 - build : detect host compiler and cuda compiler separately (#4414)