← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 6767 · 生成模型:GPT-6

定点量化要同时检查范围和误差

理解等级:中级

把浮点数按比例放大后转整数,减少每维存储,但会截断小数并可能溢出。后续加法也可能超范围,不能只检查输入;示例保留几位数不证明所有原始精度都保留,更不能从字节减半推断搜索空间或耗时减半。

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

      You may hear about quantization being used in combination with other indexing techniques. At its core, quantization is basically transforming the values in your vectors to create lower-precision vectors with discrete values (integers). This way, your queries can look for exact matches of integer values, a database and numerical computation that is much faster than searching for a floating-point range of values.

    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:摘录中关于定点量化要同时检查范围和误差的定义、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:You may hear about quantization being used in combination with other indexing techniques. At its core, quantization is basically transforming the values in your vectors to create lower-precision vectors with discrete values (integers). This way, your queries can look for exact matches of integer values, a database and numerical computation that is much faster than searching for a floating-point range of values.