<rss xmlns:atom="http://www.w3.org/2005/Atom" version="2.0">
    <channel>
        <title>Latent Experts - 标签 - vllbc02&#39;s blogs</title>
        <link>https://blog.vllbc.top/tags/latent-experts/</link>
        <description>Latent Experts - 标签 - vllbc02&#39;s blogs</description>
        <generator>Hugo -- gohugo.io</generator><language>zh-CN</language><copyright>This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.</copyright><lastBuildDate>Sat, 06 Jun 2026 00:00:00 &#43;0000</lastBuildDate><atom:link href="https://blog.vllbc.top/tags/latent-experts/" rel="self" type="application/rss+xml" /><item>
    <title>Latent Experts</title>
    <link>https://blog.vllbc.top/latent-experts/</link>
    <pubDate>Sat, 06 Jun 2026 00:00:00 &#43;0000</pubDate>
    <author>vllbc</author>
    <guid>https://blog.vllbc.top/latent-experts/</guid>
    <description><![CDATA[<h1 id="latent-experts">Latent Experts</h1>
<blockquote>
<p>来自 Sebastian Raschka 的 <em>The Big LLM Architecture
Comparison</em>。</p>
</blockquote>
<h2 id="一句话">一句话</h2>
<p>Latent Experts = MoE 中的专家在 <strong>latent space（低维）</strong>
工作，而非原始 token embedding 空间：</p>
<ul>
<li>输入 4096 维 down-project 到 1024</li>
<li>专家在 1024 维工作</li>
<li>输出 up-project 回 4096</li>
</ul>
<p>→ 节省计算量，专家可以更”小”。</p>]]></description>
</item>
</channel>
</rss>
