Why are we still launching models without simple working python example code (or llama.cpp support)?