<?xml version="1.0" encoding="utf-8" standalone="yes"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>Sequence Generation on Nalar</title>
    <link>https://nalar.dev/tags/sequence-generation/</link>
    <description>Recent content in Sequence Generation on Nalar</description>
    <generator>Hugo</generator>
    <language>en-us</language>
    <lastBuildDate>Thu, 03 Sep 2026 00:00:00 +0700</lastBuildDate>
    <atom:link href="https://nalar.dev/tags/sequence-generation/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>Understand Beam Search for Sequence Generation</title>
      <link>https://nalar.dev/understand-beam-search-for-sequence-generation/</link>
      <pubDate>Thu, 03 Sep 2026 00:00:00 +0700</pubDate>
      <guid>https://nalar.dev/understand-beam-search-for-sequence-generation/</guid>
      <description>&lt;p&gt;A model that generates text or another sequence makes a series of local decisions. At each step, it assigns scores or probabilities to possible next tokens. The simplest decoder chooses the most likely token, appends it, and repeats.&lt;/p&gt;&#xA;&lt;p&gt;That strategy is called &lt;strong&gt;greedy decoding&lt;/strong&gt;. It is cheap and easy to understand, but an early choice that looks best by itself can lead to a worse complete sequence. Once greedy decoding commits to that choice, it cannot reconsider it.&lt;/p&gt;</description>
    </item>
  </channel>
</rss>
