<?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>Vintage Slm on Derek Hommel</title>
    <link>https://snuderek.github.io/categories/vintage-slm/</link>
    <description>Recent content in Vintage Slm on Derek Hommel</description>
    <generator>Hugo</generator>
    <language>en-us</language>
    <lastBuildDate>Wed, 15 Jul 2026 17:22:00 +0900</lastBuildDate>
    <atom:link href="https://snuderek.github.io/categories/vintage-slm/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>Vintage SLM 1: Introduction - Data and Pretraining</title>
      <link>https://snuderek.github.io/posts/vintage-slm-1/</link>
      <pubDate>Wed, 15 Jul 2026 17:22:00 +0900</pubDate>
      <guid>https://snuderek.github.io/posts/vintage-slm-1/</guid>
      <description>&lt;h2 id=&#34;why-this-project&#34;&gt;Why This Project&lt;/h2&gt;
&lt;p&gt;This is the first in a series about my attempt to train a small &amp;lsquo;vintage&amp;rsquo; language model. I wanted to tackle this because in my day job, I was mostly working on agentic systems built on proprietary model backbones, so most of the work was harness construction, tool-use development, and agent evaluation. I did experiment with basic fine-tuning for style transfer using OpenAI&amp;rsquo;s now-defunct fine-tuning API for GPT 3.5-turbo, and also conducted limited open-weight model LoRA fine-tuning using HuggingFace Deep Learning Containers on GCP, but I haven&amp;rsquo;t had the opportunity to train a model from scratch.&lt;/p&gt;</description>
    </item>
  </channel>
</rss>
