Inco Splash is an open-source inference engine built around the model and Apple silicon, up to 3× the decode speed of Ollama on an M5 Max.