vllm:基于 LLM 技术的推理与服务库项目

A high-throughput and memory-efficient inference and serving engine for LLMs