llama.cpp

llama-bench: print `-n-cpu-moe` when offloaded layers > 1

#20984

Merged