<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:content="http://purl.org/rss/1.0/modules/content/"><channel><title>Softmax on HJJJ的个人小站</title><link>https://hjjj.top/tags/softmax/</link><description>Recent content in Softmax on HJJJ的个人小站</description><generator>Hugo</generator><language>en-us</language><lastBuildDate>Mon, 03 Aug 2026 00:00:00 +0800</lastBuildDate><atom:link href="https://hjjj.top/tags/softmax/index.xml" rel="self" type="application/rss+xml"/><item><title>Softmax 与缩放点积的数学：Attention 背后的概率与数值</title><link>https://hjjj.top/posts/computer-science/ai-math/softmax-and-scaled-dot-product/</link><pubDate>Mon, 03 Aug 2026 00:00:00 +0800</pubDate><guid>https://hjjj.top/posts/computer-science/ai-math/softmax-and-scaled-dot-product/</guid><description>Softmax 的定义、梯度与数值稳定性，以及 QK^T 为何要除以 sqrt(d_k)</description></item></channel></rss>