Sure, but does llama-cpp support that?