llama.cpp

c6807b3f - ci : add ubuntu cuda build, build with one arch on windows (#10456)