知识卡片
动态检索按时效和风险决策
可根据模型可知范围、信息时效、任务和成本决定是否检索。热门实体只是可能掌握的代理,不能保证当前事实正确;小于7B一律检索或大模型可免检索都不是通用规则。
这张卡片看不懂?
这张卡片已通过人工审查,不再接受失败反馈。
已有反馈 · 0 条
还没有人提交反馈。
在整书笔记中的位置
- · 检索增强生成
原书出处 · 1 处
- · 查看原书依据
Deciding When to Retrieve For each step in an agentic workflow, the LLM can advance its task using one of the following steps: Use its internal capabilities Choose from among several data stores Choose from among several software tools There can be tasks that the LLM can fully solve using its parametric memory, but one or more data stores may also contain the requisite data
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:摘录中关于动态检索按时效和风险决策的定义、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:Deciding When to Retrieve For each step in an agentic workflow, the LLM can advance its task using one of the following steps: Use its internal capabilities Choose from among several data stores Choose from among several software tools There can be tasks that the LLM can fully solve using its parametric memory, but one or more data stores may also contain the requisite data