<?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>扩散语言模型 on Docs</title>
    <link>https://docs.keinmal.top/notes/dlm/</link>
    <description>Recent content in 扩散语言模型 on Docs</description>
    <generator>Hugo --0.165.0</generator>
    <language>zh-cn</language><managingEditor>wyz2016zxc@outlook.com(Hiraeth)</managingEditor><webMaster>wyz2016zxc@outlook.com(Hiraeth)</webMaster><lastBuildDate>Thu, 08 Oct 2026 19:25:46 +0800</lastBuildDate><atom:link href="https://docs.keinmal.top/notes/dlm/index.xml" rel="self" type="application/rss+xml" /><item>
      <title>Qwen 架构</title>
      <link>https://docs.keinmal.top/notes/dlm/04-qwen-architecture/</link>
      <pubDate>Thu, 08 Oct 2026 00:00:00 +0000</pubDate><author>wyz2016zxc@outlook.com(Hiraeth)</author><guid>https://docs.keinmal.top/notes/dlm/04-qwen-architecture/</guid>
      <description>Qwen2 / Qwen3 的实测配置差异与三个坑</description></item><item>
      <title>DLM 与 A2D</title>
      <link>https://docs.keinmal.top/notes/dlm/03-dlm-a2d/</link>
      <pubDate>Thu, 08 Oct 2026 00:00:00 +0000</pubDate><author>wyz2016zxc@outlook.com(Hiraeth)</author><guid>https://docs.keinmal.top/notes/dlm/03-dlm-a2d/</guid>
      <description>扩散语言模型的数学形式、训练四步、采样控制与 A2D 转换</description></item><item>
      <title>Vocab / Tokenizer / Mask Token</title>
      <link>https://docs.keinmal.top/notes/dlm/02-vocab-tokenizer-mask/</link>
      <pubDate>Thu, 08 Oct 2026 00:00:00 +0000</pubDate><author>wyz2016zxc@outlook.com(Hiraeth)</author><guid>https://docs.keinmal.top/notes/dlm/02-vocab-tokenizer-mask/</guid>
      <description>词表大小、分词器，以及 DLM 的 mask token 三种方案</description></item><item>
      <title>Pretrain / SFT / RL</title>
      <link>https://docs.keinmal.top/notes/dlm/01-pretrain-sft-rl/</link>
      <pubDate>Thu, 08 Oct 2026 00:00:00 +0000</pubDate><author>wyz2016zxc@outlook.com(Hiraeth)</author><guid>https://docs.keinmal.top/notes/dlm/01-pretrain-sft-rl/</guid>
      <description>三个训练阶段的区别、数据格式与 rollout 顺序</description></item></channel>
</rss>
