Comment by Carrok

This is probably true, but unlike every Nvidia product we tried, it did, you know, reply to inference requests with actual output. That said, you can serve vLLM with Ray Serve. https://docs.ray.io/en/latest/serve/tutorials/vllm-example.h...