知识卡片
查询与文档改写分摊成本不同
请求时改写查询易接入,却增加在线延迟;预先为文档生成问题或上下文说明,可将成本放到入库阶段。文档变化需要重新处理,生成描述也可能错误,应保留原文和改写版本。
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
doc2query and contextual retrieval. While document rewriting techniques initially have a large cost if the data stores are very large, they can reduce latency during inference time as no or little query rewriting needs to be performed. On the other hand, query rewriting techniques are simple to implement and integrate into a RAG workflow.
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:摘录中关于查询与文档改写分摊成本不同的定义、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:doc2query and contextual retrieval. While document rewriting techniques initially have a large cost if the data stores are very large, they can reduce latency during inference time as no or little query rewriting needs to be performed. On the other hand, query rewriting techniques are simple to implement and integrate into a RAG workflow.