← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 6510.c · 生成模型:GPT-6

接受修订是代理分策略

理解等级:中级

本章规则在修订评分大于或等于当前分时接受,包括同分替换。它保证的至多是所用有限代理分不下降,不能保证数学答案更对。评分器改变后,接受路径和最终结果也会改变,需保存评分版本。

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

      if revised_score >= current_score: current_full = revised_full current_extracted = revised_extracted current_score = revised_score return { "final_full": current_full, "final_extracted": current_extracted, "steps": steps, } #A Initial response (draft) #B Run for one or more iterations #C Critique the response #D Refine the response #E Log the results #F Accept revised response if it's not worse The self_refinement_loop function runs the self-refinement procedure from the previous section for

    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:摘录中关于接受修订是代理分策略的定义、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:if revised_score >= current_score: current_full = revised_full current_extracted = revised_extracted current_score = revised_score return { "final_full": current_full, "final_extracted": current_extracted, "steps": steps, } #A Initial response (draft) #B Run for one or more iterations #C Critique the response #D Refine the response #E Log the results #F Accept revised response if it's not worse The self_refinement_loop function runs the self-refinement procedure from the previous section for