Notes 技术手记on ML & infra

机器学习、基础设施，和平时学到的东西。Machine learning, infrastructure, and whatever I learned along the way.

#ML#推理

2026

2026.10.11 从注意力公式出发，理解 KV Cache Understanding the KV Cache, Starting from the Attention Formula 生成每一个新 token 时，模型其实一直在重复计算已经算过的东西。KV Cache 做的事情只有一件：把能复用的留下来。 Every time it generates a new token, the model keeps redoing work it has already done. The KV cache does exactly one thing — it keeps what can be reused. #ML #推理