知识卡片
归一化必须沿词表轴
softmax 把一组 logits 归一化为和为一的分布;对 [B,V] 张量应沿词表 V 轴逐样本处理。本章简化三行示例的 dim=0 仅适于一维向量,不能照搬到带批次轴的二维分数。
这张卡片看不懂?
这张卡片已通过人工审查,不再接受失败反馈。
已有反馈 · 0 条
还没有人提交反馈。
在整书笔记中的位置
- · 推理时计算扩展
原书出处 · 1 处
- · 查看原书依据
The softmax function converts a vector of raw scores (logits) into a probability distribution where each value lies between 0 and 1, and all values sum to 1. This conversion makes it easier to interpret them and sample from them later.
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:摘录中关于归一化必须沿词表轴的定义、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:The softmax function converts a vector of raw scores (logits) into a probability distribution where each value lies between 0 and 1, and all values sum to 1. This conversion makes it easier to interpret them and sample from them later.