← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 6766.d · 生成模型:GPT-6

图索引提速不意味着更省内存

理解等级:中级

HNSWFlat保存原向量还要保存邻接关系,通常比单纯Flat多占空间。它的主要收益是减少查询比较,压缩是另一选择;不能把索引名换成HNSW就当作降低内存的保证,应测总常驻与构建峰值。

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

      #1 Use faiss_index_factory_str="HNSW" here if you want to reduce the RAM required. The FAISSDocumentStore in Haystack gives you three of these indexing approaches to choose from. The default 'Flat' index will give you the most accurate results (highest recall rate) but will use a lot of RAM and CPU. If you’re really constrained on RAM or CPU, like when you’re hosting your app on Hugging

    参考来源

    - 位置:《Natural Language Processing in Action, Second Edition》第10章《真实世界中的大语言模型》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Natural Language Processing in Action, Second Edition/_epub-src/OEBPS/Text/chapter-10.html) - 结论依据:摘录中关于图索引提速不意味着更省内存的定义、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:#1 Use faiss_index_factory_str="HNSW" here if you want to reduce the RAM required. The FAISSDocumentStore in Haystack gives you three of these indexing approaches to choose from. The default 'Flat' index will give you the most accurate results (highest recall rate) but will use a lot of RAM and CPU. If you’re really constrained on RAM or CPU, like when you’re hosting your app on Hugging