<?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>GLA on Hellokitty&#39;s Blog</title>
    <link>https://yangyang233333.github.io/tags/gla/</link>
    <description>Recent content in GLA on Hellokitty&#39;s Blog</description>
    <generator>Hugo</generator>
    <language>zh-CN</language>
    <lastBuildDate>Mon, 28 Sep 2026 11:04:00 +0800</lastBuildDate>
    <atom:link href="https://yangyang233333.github.io/tags/gla/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>线性注意力机制：从矩阵重排到递推记忆</title>
      <link>https://yangyang233333.github.io/posts/linear-attention-principles/</link>
      <pubDate>Mon, 28 Sep 2026 11:04:00 +0800</pubDate>
      <guid>https://yangyang233333.github.io/posts/linear-attention-principles/</guid>
      <description>从核函数与矩阵乘法结合律推导线性注意力，解释固定大小的递推状态、分块并行训练，以及 GLA、DeltaNet、Gated DeltaNet 如何管理有限记忆，并用 NumPy 验证等价关系。</description>
    </item>
  </channel>
</rss>
