<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>wxgmjfhy</title><link>https://wxgmjfhy-blog.pages.dev/</link><description>记录学习、踩坑与思考</description><generator>Hugo 0.147.0</generator><language>zh-CN</language><lastBuildDate>Tue, 20 Oct 2026 14:22:21 +0000</lastBuildDate><atom:link href="https://wxgmjfhy-blog.pages.dev/index.xml" rel="self" type="application/rss+xml"/><item><title>RL 异步 rollout</title><link>https://wxgmjfhy-blog.pages.dev/posts/rl-async-rollout/</link><pubDate>Tue, 20 Oct 2026 17:00:00 +0800</pubDate><guid>https://wxgmjfhy-blog.pages.dev/posts/rl-async-rollout/</guid><description>从 RL 基础、IMPALA/V-trace、Partial Rollout、AReaL，到 DeepSeek V4.1 / GLM-5.3 / MiMo V2.6 / Kimi K3 的异步 rollout 方案调研与总结。</description></item><item><title>关于</title><link>https://wxgmjfhy-blog.pages.dev/about/</link><pubDate>Tue, 20 Oct 2026 16:00:00 +0800</pubDate><guid>https://wxgmjfhy-blog.pages.dev/about/</guid><description>关于这个博客和我</description></item></channel></rss>