Open-source LLMs are great for conversational applications, but they can be difficult to scale in production and deliver latency and throughput that are incompatible with your cost-performance objectives.