知识卡片

普通读书笔记卡 · 8689.b · 生成模型:GPT-6

非对称量化引入零点

理解等级:高级

非对称量化用区间上下限映射整数码范围,并保存scale及zero-point。适合偏斜分布不代表总比对称好;实际零点可需取整和裁剪,比较时要含反量化误差及执行成本。

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

      Asymmetric Quantization In this setting, the 0 value in the original format is not guaranteed to be mapped to the 0 value in the integer representation. A common technique is to take the minimum and maximum value that we need represented and map it to the minimum (–127) and maximum (127) values that can be represented in int8, respectively. For example, if the range of

    参考来源

    - 位置:《Designing Large Language Model Applications》第09章《推理优化》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Designing Large Language Model Applications/_epub-src/OEBPS/ch09.html) - 结论依据:摘录中关于非对称量化引入零点的定义、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:Asymmetric Quantization In this setting, the 0 value in the original format is not guaranteed to be mapped to the 0 value in the integer representation. A common technique is to take the minimum and maximum value that we need represented and map it to the minimum (–127) and maximum (127) values that can be represented in int8, respectively. For example, if the range of