知识卡片

普通读书笔记卡 · 8704.a · 生成模型:GPT-6

多层表示支持不同检索入口

理解等级:中级

句、段、小节和整文可分别建向量,高层也可表示摘要。可先粗后细或直接查指定层,但要保存层级与来源映射;摘要表示可能丢失细节,不能拿高层相似替代事实定位。

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

      Multi-Level Embeddings If your retrieval performance requirements are stringent, a good strategy is to use multiple levels of embeddings if the cost justifies it. As an example, you can have sentence embeddings, paragraph or dialog-turn embeddings, section/subsection embeddings, or even document embeddings. The higher-level embeddings can represent the summary of the text and not necessarily the verbatim text itself. You can use different embedding models

    参考来源

    - 位置:《Designing Large Language Model Applications》第11章《表示学习与嵌入》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Designing Large Language Model Applications/_epub-src/OEBPS/ch11.html) - 结论依据:摘录中关于多层表示支持不同检索入口的定义、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:Multi-Level Embeddings If your retrieval performance requirements are stringent, a good strategy is to use multiple levels of embeddings if the cost justifies it. As an example, you can have sentence embeddings, paragraph or dialog-turn embeddings, section/subsection embeddings, or even document embeddings. The higher-level embeddings can represent the summary of the text and not necessarily the verbatim text itself. You can use different embedding models