← 返回起始笔记位置

知识卡片

普通读书笔记卡 · 6500.p.2 · 生成模型:GPT-6

候选最优由评分器定义

理解等级:中级

Best-of-N 为多个候选打分并选最高分,自一致性则按答案频率选择。前者不需要相同短答案重复出现,但效果取决于评分器是否偏向真正有用的内容;更高代理分不自动意味着更高正确率。

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

      Exercise 5.2: Using the heuristic scorer in a Best-of-N setup Modify the self-consistency implementation so that the final answer is chosen using the heuristic scorer rather than majority voting. Generate N candidate answers for each problem (where N=2 or higher), score each candidate with the heuristic scorer, and select the one with the highest score as the final prediction. (If we use a scorer instead

    参考来源

    - 位置:《Build a Reasoning Model (From Scratch)》第05章《自我修正与推理时扩展》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Build a Reasoning Model (From Scratch)/_epub-src/OEBPS/Text/chapter-5.html) - 结论依据:摘录中关于候选最优由评分器定义的定义、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:Exercise 5.2: Using the heuristic scorer in a Best-of-N setup Modify the self-consistency implementation so that the final answer is chosen using the heuristic scorer rather than majority voting. Generate N candidate answers for each problem (where N=2 or higher), score each candidate with the heuristic scorer, and select the one with the highest score as the final prediction. (If we use a scorer instead