<?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>Embedding-Scaling on Cognaptus</title>
    <link>https://cognaptus.com/tags/embedding-scaling/</link>
    <description>Recent content in Embedding-Scaling on Cognaptus</description>
    <generator>Hugo -- 0.145.0</generator>
    <language>en-us</language>
    <lastBuildDate>Fri, 11 Sep 2026 00:00:00 +0000</lastBuildDate>
    <atom:link href="https://cognaptus.com/tags/embedding-scaling/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>When the Next Sparse Billion Should Go to Memory, Not Experts</title>
      <link>https://cognaptus.com/blog/2026-09-11-when-the-next-sparse-billion-should-go-to-memory-not-experts/</link>
      <pubDate>Fri, 11 Sep 2026 00:00:00 +0000</pubDate>
      <guid>https://cognaptus.com/blog/2026-09-11-when-the-next-sparse-billion-should-go-to-memory-not-experts/</guid>
      <description>LongCat-Flash experiments suggest that once MoE expert scaling reaches a high-sparsity regime, some marginal capacity may be better allocated to sparse lookup memory—but only within tight architectural and serving constraints.</description>
    </item>
  </channel>
</rss>
