#architecture

Every LLM request has two distinct phases with different performance characteristics. Understanding them is the key to targeted optimization.