vLLM large scale serving: DeepSeek 2.2k tok/s/h200 with wide-epblog.vllm.ai
98 points by robertnishihara 17 hours ago | 12 comments