知识卡片

普通读书笔记卡 · 8703.c · 生成模型:GPT-6

滑动窗口以重复换邻近上下文

理解等级:中级

重叠切块让相邻段落的关键信息共同出现,降低边界割裂风险。重叠会增加存储和重复召回,且无法覆盖任意远程依赖;需要去重和窗口大小评估。

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

      Sliding Window Chunking Consider a situation where the embedding similarity function returns a unit in Chunk 45 as the most similar vector to your query vector. However, text in Chunk 44, which immediately precedes Chunk 45 in the document, contains relevant information contextualizing Chunk 45. The vectors in Chunk 44 have a very low similarity score with the query, and as a result, Chunk 44

    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:摘录中关于滑动窗口以重复换邻近上下文的定义、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:Sliding Window Chunking Consider a situation where the embedding similarity function returns a unit in Chunk 45 as the most similar vector to your query vector. However, text in Chunk 44, which immediately precedes Chunk 45 in the document, contains relevant information contextualizing Chunk 45. The vectors in Chunk 44 have a very low similarity score with the query, and as a result, Chunk 44