vLLM is a high-throughput and memory-efficient inference engine for large language models, designed...