知识卡片

普通读书笔记卡 · 6509.d · 生成模型:GPT-6

对数概率越大越不负

理解等级:初级

概率在零到一之间,其自然对数不大于零。比较同一评分协议下的对数概率时,更接近零表示模型赋予更高概率;不能因为绝对值更大就误判更好,也不能把这种高概率解读为事实正确率。

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

      values closer to zero are better, since zero corresponds to a probability of one, while very negative values correspond to extremely small probabilities. This makes it easy to compare tokens: the least negative log-probability is the most likely one, and the most negative log-probability is the least likely.

    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:摘录中关于对数概率越大越不负的定义、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:values closer to zero are better, since zero corresponds to a probability of one, while very negative values correspond to extremely small probabilities. This makes it easy to compare tokens: the least negative log-probability is the most likely one, and the most negative log-probability is the least likely.