Open-source Large Language Models (LLMs) have become much easier to run. But one thing is often misunderstood: Running a model locally is not the same as serving a model efficiently.