llm-d v1.0: State-of-the-art inference