Tag
A step-by-step, first-principles explainer on how prefill and decode actually work inside LLM inference, starting from how ChatGPT-style chats process prompts via tokenization and autoregressive token generation.