Stop guessing your llama.cpp config. Give it a target tok/s and a quality budget, and it…