<?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>LLM训练 on caulif的个人博客</title><link>https://www.caulif.com/tags/llm%E8%AE%AD%E7%BB%83/</link><description>Recent content in LLM训练 on caulif的个人博客</description><generator>Hugo -- 0.154.5</generator><language>zh-cn</language><lastBuildDate>Thu, 09 Jul 2026 00:00:00 +0800</lastBuildDate><atom:link href="https://www.caulif.com/tags/llm%E8%AE%AD%E7%BB%83/index.xml" rel="self" type="application/rss+xml"/><item><title>AI 时代的思维框架读后小记</title><link>https://www.caulif.com/posts/ai%E6%97%B6%E4%BB%A3%E7%9A%84%E6%80%9D%E7%BB%B4%E6%A1%86%E6%9E%B6%E8%AF%BB%E5%90%8E%E5%B0%8F%E8%AE%B0/</link><pubDate>Thu, 09 Jul 2026 00:00:00 +0800</pubDate><guid>https://www.caulif.com/posts/ai%E6%97%B6%E4%BB%A3%E7%9A%84%E6%80%9D%E7%BB%B4%E6%A1%86%E6%9E%B6%E8%AF%BB%E5%90%8E%E5%B0%8F%E8%AE%B0/</guid><description>记录一个关于推理与训练的地形类比：模型参数像初始地形，prompt 影响局部路径，训练则在长期上塑造地貌与变化规则。</description></item><item><title>反向传播（Backpropagation）学习笔记</title><link>https://www.caulif.com/posts/%E5%8F%8D%E5%90%91%E4%BC%A0%E6%92%AD%E5%AD%A6%E4%B9%A0%E7%AC%94%E8%AE%B0/</link><pubDate>Fri, 26 Jun 2026 00:00:00 +0800</pubDate><guid>https://www.caulif.com/posts/%E5%8F%8D%E5%90%91%E4%BC%A0%E6%92%AD%E5%AD%A6%E4%B9%A0%E7%AC%94%E8%AE%B0/</guid><description>整理反向传播的 Loss、梯度、链式法则、优化器更新，以及 LLM 训练中常见的显存优化方法。</description></item><item><title>长上下文退化</title><link>https://www.caulif.com/posts/%E4%B8%8A%E4%B8%8B%E6%96%87%E9%80%80%E5%8C%96/</link><pubDate>Fri, 26 Jun 2026 00:00:00 +0800</pubDate><guid>https://www.caulif.com/posts/%E4%B8%8A%E4%B8%8B%E6%96%87%E9%80%80%E5%8C%96/</guid><description>整理长上下文退化的含义、位置偏置、长距离泛化、训练分布、检索与推理差异，以及 Agent 场景中的 context rot。</description></item></channel></rss>