Decision models now run on-device in llama.cpp

A post says decision models now run on-device in llama.cpp, describing the option as free, fast and private. It shares the command llama serve -hf ggml-org/Kev-4B-GGUF.