← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 8706.e · 生成模型:GPT-6

复杂检索可拆成带依赖子问

理解等级:中级

将多跳问题拆成可检索子问,独立子问可并行,依赖上一步结果的子问需顺序执行。应保存子问与原问题关系,避免分别得到正确片段却错误连接实体、时间或因果关系。

这张卡片看不懂?

已有反馈 · 0 条

    还没有人提交反馈。

    在整书笔记中的位置

    原书出处 · 1 处

    • · 查看原书依据

      query decomposition. For complex queries in an agentic workflow, we can have the LLM divide the task into multiple queries that can be executed sequentially or in parallel, depending on how the query was decomposed. We discussed query decomposition techniques in Chapter 10.

    参考来源

    - 位置:《Designing Large Language Model Applications》第12章《检索增强生成》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Designing Large Language Model Applications/_epub-src/OEBPS/ch12.html) - 结论依据:摘录中关于复杂检索可拆成带依赖子问的定义、比较或条件支撑卡片;原书表述不严谨的范围已在覆盖记录逐项限定,不把案例结果外推为普遍结论。 - 原始内容:query decomposition. For complex queries in an agentic workflow, we can have the LLM divide the task into multiple queries that can be executed sequentially or in parallel, depending on how the query was decomposed. We discussed query decomposition techniques in Chapter 10.