<?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>Variable-Length on Tenu Tech Brief</title>
    <link>https://cluster-site.onrender.com/tags/variable-length/</link>
    <description>Recent content in Variable-Length on Tenu Tech Brief</description>
    <generator>Hugo -- 0.146.0</generator>
    <language>en-us</language>
    <lastBuildDate>Tue, 24 Feb 2026 06:03:06 +0000</lastBuildDate>
    <atom:link href="https://cluster-site.onrender.com/tags/variable-length/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>Speeding Up Variable-Length Training with Dynamic Context Parallelism and NVIDIA Megatron Core</title>
      <link>https://cluster-site.onrender.com/posts/speeding-up-variable-length-training-with-dynamic-context-parallelism-and-nvidia-megatron-core/</link>
      <pubDate>Wed, 28 Jan 2026 16:28:06 +0000</pubDate>
      <guid>https://cluster-site.onrender.com/posts/speeding-up-variable-length-training-with-dynamic-context-parallelism-and-nvidia-megatron-core/</guid>
      <description>• This post introduces Dynamic Context Parallelism (Dynamic-CP), a scheduling approach in NVIDIA Megatron Core used for LLM post-training or DiT pre-training. • It dynamically sele</description>
    </item>
  </channel>
</rss>
