VLLM is an open source library for fast, easy-to-use LLM inference and serving. It optimizes hundreds of language models across diverse data-center hardware—NVIDIA and AMD GPUs, Google TPUs, AWS Trainium, Intel CPUs—usin...