<?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>Attention on Hellokitty&#39;s Blog</title>
    <link>https://yangyang233333.github.io/tags/attention/</link>
    <description>Recent content in Attention on Hellokitty&#39;s Blog</description>
    <generator>Hugo</generator>
    <language>zh-CN</language>
    <lastBuildDate>Tue, 25 Aug 2026 21:00:00 +0800</lastBuildDate>
    <atom:link href="https://yangyang233333.github.io/tags/attention/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>Mini-SGLang 源码阅读（四）：Engine、Attention Backend 与 CUDA Graph</title>
      <link>https://yangyang233333.github.io/posts/mini-sglang-source-reading-engine/</link>
      <pubDate>Tue, 25 Aug 2026 21:00:00 +0800</pubDate>
      <guid>https://yangyang233333.github.io/posts/mini-sglang-source-reading-engine/</guid>
      <description>从 GPU 推理的 kernel 启动开销和 Prefill/Decode 差异出发，分析 Mini-SGLang Engine 的完整执行路径。</description>
    </item>
  </channel>
</rss>
