分块Chunking

把长文档切成小块再送进模型。切得好不好，直接决定回答准不准。Cut a long document into pieces before feeding it to the model. How you chunk decides how accurate the answer is.

RAG 流程里把文档切成段落的过程。切太小丢上下文，切太大塞不进上下文窗口。

The step in a RAG pipeline that splits a document into passages. Too small loses context; too large does not fit the context window.

它是 RAG 里最容易被忽略、却最影响效果的环节。很多「AI 答得不准」的原因是切块切坏了。The most overlooked yet most consequential step in RAG. Many "the AI answered wrongly" cases trace back to bad chunking.

去 Models 图谱：看这个模型有哪些量化版、按你的显存该下哪一档Models atlas: which quantisations exist and which tier fits your VRAM ↗

回到名词地图看第2站的其他 35 个词 →Back to the map: the other 35 terms in station 2 →

• Chroma 分块指南Chunking – Chroma Docs ↗