知识卡片
知识增强管线保留证据链
结构图:
flowchart TD
Q[原请求与约束] --> W[可选改写或分解]
W --> R[召回必要证据]
R --> K[重排和过滤]
K --> F[压缩并保留来源与限定]
F --> I[按预算编排上下文]
I --> G[生成及具体引用]
G --> V[支持性 完整性与成本验收]
V --> A[必要时主动检索或承认未知]
A --> R
各阶段需要真实输入输出记录;检索代理、压缩结果和金标训练信号不能混成在线证据。
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
rewrite-retrieve-rerank-refine-insert-generate pipeline. We highlighted the effectiveness of RAG in various scenarios, including integration of external knowledge, retrieval of past conversational history, dynamic selection of few-shot learning examples, and tool selection. We also explored the limitations of RAG and scenarios where RAG may not be effective.
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:摘录中关于知识增强管线保留证据链的定义、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:rewrite-retrieve-rerank-refine-insert-generate pipeline. We highlighted the effectiveness of RAG in various scenarios, including integration of external knowledge, retrieval of past conversational history, dynamic selection of few-shot learning examples, and tool selection. We also explored the limitations of RAG and scenarios where RAG may not be effective.