上下文窗口的代价 · 学 AI

解决什么问题What problem it solves

上下文开得越大越慢越贵，不是免费扩容。A larger context window is slower and more expensive — it is not free capacity.

这是什么What it is

窗口变大意味着注意力计算量和 KV 缓存占用同步上升。

A bigger window means attention computation and KV cache memory both grow.

为什么需要知道Why it matters

解释了一个反直觉现象。把所有资料一次性塞进去，往往不如做 RAG 分批检索。It explains a counter-intuitive result: pasting in everything at once is often worse than retrieving in batches with RAG.

接着读什么，为什么What to read next, and why

上下文窗口Context Window
补上长输入的处理量与内存代价。Account for computation and memory costs of long input.
参数Parameter
一起用于《AI 常见误区总索引》Used together in “AI Misconceptions: Index”
基准测试Benchmark
一起用于《AI 常见误区总索引》Used together in “AI Misconceptions: Index”

现在能做什么What you can do now

RAG server 的做法How a RAG server does it (context7) ↗

回到名词地图看第4站的其他 35 个词 →Back to the map: the other 35 terms in station 4 →

出处Sources

RAG server 的做法How a RAG server does it (context7) ↗

上下文窗口的代价Context Cost