<?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>PI05 on Yukari's Blog</title><link>https://yukar.icu/tags/pi05/</link><description>Recent content in PI05 on Yukari's Blog</description><generator>Hugo</generator><language>zh-cn</language><lastBuildDate>Mon, 04 Aug 2025 00:00:00 +0000</lastBuildDate><atom:link href="https://yukar.icu/tags/pi05/index.xml" rel="self" type="application/rss+xml"/><item><title>PI05 模型架构与推理流程详解</title><link>https://yukar.icu/posts/pi05modelingworkflow/</link><pubDate>Mon, 04 Aug 2025 00:00:00 +0000</pubDate><guid>https://yukar.icu/posts/pi05modelingworkflow/</guid><description>逐行拆解 PI05（Pi 0.5）模型实现，从图像/语言观测到动作序列输出的完整源码分析。覆盖模型初始化、训练前向、推理采样、权重加载、数据预处理及所有工具函数。</description></item></channel></rss>