<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>系统设计 on jimyag's Blog</title><link>https://jimyag.com/blog/%E7%B3%BB%E7%BB%9F%E8%AE%BE%E8%AE%A1/</link><description>jimyag 的个人博客，记录 Kubernetes、KubeVirt、Kube-OVN、Firecracker、Go 与云原生工程实践。</description><generator>Hugo</generator><language>zh-CN</language><copyright>Copyright © 2019, jimyag.</copyright><lastBuildDate>Tue, 27 Oct 2026 16:13:00 +0800</lastBuildDate><atom:link href="https://jimyag.com/blog/%E7%B3%BB%E7%BB%9F%E8%AE%BE%E8%AE%A1/index.xml" rel="self" type="application/rss+xml"/><item><title>大模型推理引擎原理与性能优化：从 vLLM、投机推理到 KV Cache 调度</title><link>https://jimyag.com/posts/llm-inference-engine-optimization-guide/</link><pubDate>Tue, 27 Oct 2026 16:13:00 +0800</pubDate><guid>https://jimyag.com/posts/llm-inference-engine-optimization-guide/</guid><description>从一次请求的 Prefill、Decode 和 KV Cache 开始，讲清 EAGLE3、DFlash、DSpark、拒绝采样、SJF、Weight Offload、用户级 Token/s 限流与缓存最低保留时间，并附循序渐进的技术面试问题及参考答案。</description></item></channel></rss>