vLLM is a fast and easy-to-use library for LLM inference and serving. Originally developed...