<?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>混合注意力 on AI 实战派 · 从技术到赚钱</title><link>https://guijiagi.com/tags/%E6%B7%B7%E5%90%88%E6%B3%A8%E6%84%8F%E5%8A%9B/</link><description>Recent content in 混合注意力 on AI 实战派 · 从技术到赚钱</description><generator>Hugo</generator><language>zh-cn</language><copyright>本站内容采用 CC BY-NC-SA 4.0 国际许可协议授权</copyright><lastBuildDate>Fri, 09 Oct 2026 15:00:00 +0800</lastBuildDate><atom:link href="https://guijiagi.com/tags/%E6%B7%B7%E5%90%88%E6%B3%A8%E6%84%8F%E5%8A%9B/index.xml" rel="self" type="application/rss+xml"/><item><title>DeepSeek-V4.1-Flash 架构拆解：552B MoE 背后的非对称激活与混合注意力</title><link>https://guijiagi.com/posts/2026-10-09-01-deepseek-v41-flash-architecture/</link><pubDate>Fri, 09 Oct 2026 15:00:00 +0800</pubDate><guid>https://guijiagi.com/posts/2026-10-09-01-deepseek-v41-flash-architecture/</guid><description>V4.1-Flash 用 552B 总参、输入 8B 输出 16B 的非对称激活，叠加 CSA 与 HCA 混合注意力，把百万上下文的推理成本压到新低。</description></item></channel></rss>