Ollama doesn't have their own inference engine, they just wrap llama.cpp. But yes, it will be awesome when it's more generally available.