Can we instead use deep infra API to have full scale LLM running in the back? I would love that, easy access from amethyst..