llama.cpp

9c600bcd - llama-bench: print `-n-cpu-moe` when offloaded layers > 1 (#20984)