<?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>循环模型 on AI 实战派 · 从技术到赚钱</title><link>https://guijiagi.com/tags/%E5%BE%AA%E7%8E%AF%E6%A8%A1%E5%9E%8B/</link><description>Recent content in 循环模型 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>Mon, 19 Oct 2026 14:24:00 +0800</lastBuildDate><atom:link href="https://guijiagi.com/tags/%E5%BE%AA%E7%8E%AF%E6%A8%A1%E5%9E%8B/index.xml" rel="self" type="application/rss+xml"/><item><title>英伟达低调放出 Dory1.0-4B，运行时自选一到四圈推理深度，一个权重怎么两用？</title><link>https://guijiagi.com/posts/2026-10-19-04-nvidia-dory-4b-loop-depth-runtime/</link><pubDate>Mon, 19 Oct 2026 14:24:00 +0800</pubDate><guid>https://guijiagi.com/posts/2026-10-19-04-nvidia-dory-4b-loop-depth-runtime/</guid><description>NVIDIA Nemotron Labs 发布 40 亿参数、256K 上下文的 Dory1.0-4B，推理时可在一个到四个循环之间切换而无需换检查点，四圈 SWE-bench Verified 厂商自报 61.8 分，循环注意力缓存保持单圈大小。</description></item></channel></rss>