<?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>Posts on Blog</title><link>https://jubsteven.github.io/blog/posts/</link><description>Recent content in Posts on Blog</description><generator>Hugo</generator><language>zh-cn</language><lastBuildDate>Sun, 17 May 2026 00:00:00 +0800</lastBuildDate><atom:link href="https://jubsteven.github.io/blog/posts/index.xml" rel="self" type="application/rss+xml"/><item><title>DeepSeek-V4技术报告（一）</title><link>https://jubsteven.github.io/blog/posts/deepseek-report-1/</link><pubDate>Sun, 17 May 2026 00:00:00 +0800</pubDate><guid>https://jubsteven.github.io/blog/posts/deepseek-report-1/</guid><description>从DeepSeek V4技术报告出发，梳理Speculative Decoding、MTP与DeepSeek MoE等关键机制。</description></item><item><title>Loss in Agentic RL</title><link>https://jubsteven.github.io/blog/posts/rl-loss/</link><pubDate>Sun, 10 May 2026 19:00:00 +0800</pubDate><guid>https://jubsteven.github.io/blog/posts/rl-loss/</guid><description>在前段时间基于VeRL的Search-R1仓库进行了一些Agentic Search相关的探索，也算是通过一些实践来积累了一点Agentic RL的基本常识。</description></item></channel></rss>