知识卡片

普通读书笔记卡 · 8710 · 生成模型:GPT-6

证据压缩要保持回答必要条件

理解等级:中级

精炼可摘要、改写或过滤候选以适应上下文预算。压缩需保留实体、单位、否定、时间及来源,不能仅按篇幅缩短率判断成功;短摘要丢掉限定后可能更容易造成错误结论。

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

      Refine Once the candidate texts relevant to the given query are retrieved and selected, they can be fed to the LLM. However, the LLM context window is limited, so we might want to reduce the length of the retrieved texts. We might also want to rephrase it so that it is more amenable to being processed by the LLM. Another possible operation could be to

    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:摘录中关于证据压缩要保持回答必要条件的定义、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:Refine Once the candidate texts relevant to the given query are retrieved and selected, they can be fed to the LLM. However, the LLM context window is limited, so we might want to reduce the length of the retrieved texts. We might also want to rephrase it so that it is more amenable to being processed by the LLM. Another possible operation could be to