知识卡片

普通读书笔记卡 · 6500.p.1 · 生成模型:GPT-6

长度度量决定简洁奖励含义

理解等级:中级

本章启发式按字符数给指数衰减奖励;字符数容易获得,却不等于模型 token 成本,且受语言和排版影响。用它衡量简洁应注明口径;若优化推理资源,应另记实际 token、生成时间和质量约束。

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

      For simplicity, the brevity bonus is computed using the number of characters rather than tokens, which avoids passing the tokenizer to the scoring function. Using token counts would also be reasonable (and may even be preferable).

    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:摘录中关于长度度量决定简洁奖励含义的定义、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:For simplicity, the brevity bonus is computed using the number of characters rather than tokens, which avoids passing the tokenizer to the scoring function. Using token counts would also be reasonable (and may even be preferable).