<?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>GDN on AI 实战派 · 从技术到赚钱</title><link>https://guijiagi.com/tags/gdn/</link><description>Recent content in GDN 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 15:54:00 +0800</lastBuildDate><atom:link href="https://guijiagi.com/tags/gdn/index.xml" rel="self" type="application/rss+xml"/><item><title>Qwen3.5 在 Blackwell 上跑到 25K TPS/GPU：混合架构的服务化难题</title><link>https://guijiagi.com/posts/2026-10-09-10-qwen35-gdn-25k-tps/</link><pubDate>Fri, 09 Oct 2026 15:54:00 +0800</pubDate><guid>https://guijiagi.com/posts/2026-10-09-10-qwen35-gdn-25k-tps/</guid><description>Qwen3.5 混合全注意力与 Gated Delta Network 层，vLLM 在 Blackwell 上单 GPU 冲到 25K TPS。难点在异构状态的跨 worker 传输。</description></item></channel></rss>