<?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>Temperature on Sanketh's Blog</title><link>https://sankethbk.github.io/blog/tags/temperature/</link><description>Recent content in Temperature on Sanketh's Blog</description><generator>Hugo -- 0.170.1</generator><language>en-us</language><lastBuildDate>Sat, 21 Nov 2026 00:00:00 +0530</lastBuildDate><atom:link href="https://sankethbk.github.io/blog/tags/temperature/index.xml" rel="self" type="application/rss+xml"/><item><title>Decoding Strategies: Greedy, Temperature, Top-k, Top-p, and Beam Search</title><link>https://sankethbk.github.io/blog/posts/ml/2026-11-21-5-decoding-strategies-from-logits-to-a-generated-answer/</link><pubDate>Sat, 21 Nov 2026 00:00:00 +0530</pubDate><guid>https://sankethbk.github.io/blog/posts/ml/2026-11-21-5-decoding-strategies-from-logits-to-a-generated-answer/</guid><description>&lt;h1 id="decoding-strategies-greedy-temperature-top-k-top-p-and-beam-search"&gt;Decoding Strategies: Greedy, Temperature, Top-k, Top-p, and Beam Search&lt;/h1&gt;
&lt;p&gt;The &lt;a href="https://sankethbk.github.io/blog/posts/ml/2026-11-10-4-kv-cache-making-autoregressive-generation-practical"&gt;KV-cache note&lt;/a&gt; made the generation loop practical: prefill the prompt, keep each layer&amp;rsquo;s keys and values, and run only the newest token through the model at every decode step. But it deliberately skipped the final decision in that loop.&lt;/p&gt;
&lt;p&gt;After the final Transformer layer and language-model head, we have one logit per vocabulary token. Which token do we actually put into the answer?&lt;/p&gt;</description></item></channel></rss>