<?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>Hyperparameter-Transfer on Cognaptus</title>
    <link>https://cognaptus.com/tags/hyperparameter-transfer/</link>
    <description>Recent content in Hyperparameter-Transfer on Cognaptus</description>
    <generator>Hugo -- 0.145.0</generator>
    <language>en-us</language>
    <lastBuildDate>Wed, 23 Sep 2026 00:00:00 +0000</lastBuildDate>
    <atom:link href="https://cognaptus.com/tags/hyperparameter-transfer/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>Before You Spend 10 Trillion Tokens: Separate Width From Horizon</title>
      <link>https://cognaptus.com/blog/2026-09-23-before-you-spend-10-trillion-tokens-separate-width-from-horizon/</link>
      <pubDate>Wed, 23 Sep 2026 00:00:00 +0000</pubDate>
      <guid>https://cognaptus.com/blog/2026-09-23-before-you-spend-10-trillion-tokens-separate-width-from-horizon/</guid>
      <description>A two-step learning-rate transfer method shows how MoE training teams can reduce full-scale tuning compute by separating model-width transfer from token-horizon extrapolation.</description>
    </item>
  </channel>
</rss>
