← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 8702.e · 生成模型:GPT-6

乘积量化保存子空间码本索引

理解等级:高级

PQ把向量切成子空间,在每个子空间训练码本并用最近中心编号表示。每段256中心时通常每段占1字节,比较可借助码本查表;编号数量不是重新学到的语义维度,码本存储也不能忽略。

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

      Product Quantization Another promising quantization method is called product quantization. In this technique, a vector is divided into chunks of equal size. The chunks are then clustered. The number of clusters is set to the number of values that can be represented by the quantized embedding. For example, if we aim to quantize to int8, then the number of values that can be represented is

    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:摘录中关于乘积量化保存子空间码本索引的定义、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:Product Quantization Another promising quantization method is called product quantization. In this technique, a vector is divided into chunks of equal size. The chunks are then clustered. The number of clusters is set to the number of values that can be represented by the quantized embedding. For example, if we aim to quantize to int8, then the number of values that can be represented is