llm 2 Prompt Caching: Why Does Changing One Token Break Reuse After It? Oct 12, 2026 What Happens in an LLM Call? From Requests and Sampling to Structured Responses Aug 14, 2026