<?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>LLM on Kin Ian Lo</title>
    <link>https://kinianlo.github.io/tags/llm/</link>
    <description>Recent content in LLM on Kin Ian Lo</description>
    <image>
      <title>Kin Ian Lo</title>
      <url>https://kinianlo.github.io/images/papermod-cover.png</url>
      <link>https://kinianlo.github.io/images/papermod-cover.png</link>
    </image>
    <generator>Hugo -- 0.127.0</generator>
    <language>en</language>
    <lastBuildDate>Mon, 17 Jun 2024 01:06:04 +0000</lastBuildDate>
    <atom:link href="https://kinianlo.github.io/tags/llm/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>Another derivation for &#34;Obtaining logprobs from an LLM API&#34;</title>
      <link>https://kinianlo.github.io/posts/llm-api-probs/</link>
      <pubDate>Mon, 17 Jun 2024 01:06:04 +0000</pubDate>
      <guid>https://kinianlo.github.io/posts/llm-api-probs/</guid>
      <description>OpenAI has modified their API to return the log probabilities before any logit bias is applied. Hence the methods described in this article are no longer applicable.
The aim of this article is to provide an alternative derivation of the results in Mattew Finlayson&amp;rsquo;s article Obtaining logprobs from an LLM API.
Most LLM APIs return the logprobs of only the top-$ k $ predictions. The value of $ k $ is often small, in the order of 10.</description>
    </item>
  </channel>
</rss>
