<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>Training - Developers Digest</title>
    <link>https://www.developersdigest.tech/blog/tags/training</link>
    <description>Articles about Training on Developers Digest</description>
    <language>en</language>
    <lastBuildDate>Sun, 06 Sep 2026 01:15:51 GMT</lastBuildDate>
    <atom:link href="https://www.developersdigest.tech/blog/tags/training/feed.xml" rel="self" type="application/rss+xml" />
    <item>
      <title><![CDATA[Distilling an LLM on One GPU: Offline Top-K Logits and a Fused Chunked KL Loss]]></title>
      <link>https://www.developersdigest.tech/blog/efficient-llm-distillation-single-gpu-2026</link>
      <guid isPermaLink="true">https://www.developersdigest.tech/blog/efficient-llm-distillation-single-gpu-2026</guid>
      <description><![CDATA[Multiverse Computing open-sources two changes that make knowledge distillation cheap enough to run at scale: caching the teacher's top-100 logits once so it never sits in memory during training, and a fused chunked KL loss that never materializes the vocab-by-sequence matrix. A GPT-OSS 20B distillation at 32K context drops from four GPU nodes to one, with step time down 5x.]]></description>
      <pubDate>Mon, 10 Aug 2026 00:00:00 GMT</pubDate>
      <category>News</category>
      <category>LLMs</category>
      <category>Open Source</category>
      <category>Training</category>
      <category>GPU</category>
      <enclosure url="https://www.developersdigest.tech/images/blog/agent-memory-benchmarks-not-enough/hero.webp" type="image/webp" />
    </item>
  </channel>
</rss>