<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Pretrain on Jasaxion一只大雄</title><link>https://jasaxion.github.io/keywords/pretrain/</link><description>Recent content in Pretrain on Jasaxion一只大雄</description><generator>Hugo</generator><language>zh-CN</language><lastBuildDate>Mon, 28 Sep 2026 00:00:00 +0000</lastBuildDate><atom:link href="https://jasaxion.github.io/keywords/pretrain/index.xml" rel="self" type="application/rss+xml"/><item><title>[基模杂谈]代码预训练数据管线经验杂谈</title><link>https://jasaxion.github.io/posts/2026/09/%E5%9F%BA%E6%A8%A1%E6%9D%82%E8%B0%88%E4%BB%A3%E7%A0%81%E9%A2%84%E8%AE%AD%E7%BB%83%E6%95%B0%E6%8D%AE%E7%AE%A1%E7%BA%BF%E7%BB%8F%E9%AA%8C%E6%9D%82%E8%B0%88/</link><pubDate>Mon, 28 Sep 2026 00:00:00 +0000</pubDate><guid>https://jasaxion.github.io/posts/2026/09/%E5%9F%BA%E6%A8%A1%E6%9D%82%E8%B0%88%E4%BB%A3%E7%A0%81%E9%A2%84%E8%AE%AD%E7%BB%83%E6%95%B0%E6%8D%AE%E7%AE%A1%E7%BA%BF%E7%BB%8F%E9%AA%8C%E6%9D%82%E8%B0%88/</guid><description>&lt;blockquote>
&lt;p>随着蒸馏实验的不断发展，现在越来越多的研究发现数据产生的价值越大越模型架构等，把数据做好模型能力有稳步的提升。代码数据预训练的来源和流程，优化路径包括：去重、模型打分、合成数据与过程数据融合，助力模型性能提升。&lt;/p></description></item></channel></rss>