How to Serve an LLM with an OpenAI-Compatible API on a Dedicated GPU Server