知识卡片

普通读书笔记卡 · 6508.e · 生成模型:GPT-6

提前停止要相对总预算证明

理解等级:高级

固定最大候选数 N 时,某答案票数已超过 N/2,后续所有票也不能改变其严格多数,可安全提前结束该规则。仅超过当前已采样数的一半通常不足以保证最终赢家;提前停止策略还要计入无效票与既定平票规则。

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

      implement this simple tie-breaking rule. Exercise 4.4: Early stopping in self-consistency sampling To improve computational efficiency, implement an early-stopping version of self-consistency that ends sampling once more than half of the answers agree. Choosing temperature and top-p settings For reasonable results, temperature settings are usually between 0.5 and 0.9, and a top-p value of 0.7 to 0.9 is typical. As a rule of thumb, if

    参考来源

    - 位置:《Build a Reasoning Model (From Scratch)》第04章《推理时计算扩展》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Build a Reasoning Model (From Scratch)/_epub-src/OEBPS/Text/chapter-4.html) - 结论依据:摘录中关于提前停止要相对总预算证明的定义、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:implement this simple tie-breaking rule. Exercise 4.4: Early stopping in self-consistency sampling To improve computational efficiency, implement an early-stopping version of self-consistency that ends sampling once more than half of the answers agree. Choosing temperature and top-p settings For reasonable results, temperature settings are usually between 0.5 and 0.9, and a top-p value of 0.7 to 0.9 is typical. As a rule of thumb, if