llama.cpp

8c851fcc - server: support autoload model, support preset-only options