<?xml version="1.0" encoding="utf-8" standalone="yes"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>Context Window on Nalar</title>
    <link>https://nalar.dev/tags/context-window/</link>
    <description>Recent content in Context Window on Nalar</description>
    <generator>Hugo</generator>
    <language>en-us</language>
    <lastBuildDate>Sat, 12 Sep 2026 00:00:00 +0700</lastBuildDate>
    <atom:link href="https://nalar.dev/tags/context-window/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>Extend RoPE Context Windows with Position Interpolation</title>
      <link>https://nalar.dev/extend-rope-context-windows-with-position-interpolation/</link>
      <pubDate>Sat, 12 Sep 2026 00:00:00 +0700</pubDate>
      <guid>https://nalar.dev/extend-rope-context-windows-with-position-interpolation/</guid>
      <description>&lt;h1 id=&#34;extend-rope-context-windows-with-position-interpolation&#34;&gt;Extend RoPE Context Windows with Position Interpolation&lt;/h1&gt;&#xA;&lt;p&gt;A RoPE-based language model trained on sequences up to a fixed length can behave poorly when inference suddenly asks it to process much larger position indices. The tokens are valid, but the positional pattern can move far outside the range used during training.&lt;/p&gt;&#xA;&lt;p&gt;&lt;strong&gt;Position interpolation&lt;/strong&gt; changes that geometry. Instead of sending larger position indices directly into rotary position embeddings, it compresses a longer sequence into the positional range the model already uses. With suitable adaptation, this can extend the usable context window without changing the transformer architecture.&lt;/p&gt;</description>
    </item>
  </channel>
</rss>
