知识卡片
分文档生成再融合有额外风险
逐篇文档生成后组合,能减少单次上下文长度,却增加调用成本并可能错过跨文档关系。融合时需处理矛盾、重复和各答案的证据,不能直接多数投票决定真实事实。
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
The standard approach is to stuff all the content, or at least as much as can fit, into the context window. An alternative is to feed each retrieved document/summary/note prefixed to the input separately to the LLM, and then combine the outputs.
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:摘录中关于分文档生成再融合有额外风险的定义、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:The standard approach is to stuff all the content, or at least as much as can fit, into the context window. An alternative is to feed each retrieved document/summary/note prefixed to the input separately to the LLM, and then combine the outputs.