llama.cpp

fix: use `vm_allocate` to allocate CPU backend buffer on macOS

#9875

Merged