<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>工程 on 墨然</title><link>https://moran.is-a.dev/tags/%E5%B7%A5%E7%A8%8B/</link><description>Recent content in 工程 on 墨然</description><generator>Hugo</generator><language>zh-cn</language><lastBuildDate>Sun, 30 Nov 2025 14:06:00 +0800</lastBuildDate><atom:link href="https://moran.is-a.dev/tags/%E5%B7%A5%E7%A8%8B/atom.xml" rel="self" type="application/rss+xml"/><item><title>推理服务别只盯模型：我踩坑后总结的三件小事</title><link>https://moran.is-a.dev/posts/llm-serving-basics/</link><pubDate>Sun, 30 Nov 2025 14:06:00 +0800</pubDate><guid>https://moran.is-a.dev/posts/llm-serving-basics/</guid><description>用户觉得“模型不稳定”，很多时候是网关、队列、超时策略在暗地里打架。</description></item><item><title>上下文窗口这事儿：我怎么让大模型“别忘太快”</title><link>https://moran.is-a.dev/posts/llm-context-window/</link><pubDate>Fri, 12 Sep 2025 10:05:00 +0800</pubDate><guid>https://moran.is-a.dev/posts/llm-context-window/</guid><description>我以前总以为模型“记性差”，后来才发现：很多遗忘是我自己喂的内容太乱。</description></item></channel></rss>