<?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>KV缓存 on zo0043</title><link>http://blog.zero43.top/tags/kv%E7%BC%93%E5%AD%98/</link><description>Recent content in KV缓存 on zo0043</description><image><title>zo0043</title><url>https://i.postimg.cc/7hwBy7VS/calcr.png</url><link>https://i.postimg.cc/7hwBy7VS/calcr.png</link></image><generator>Hugo -- 0.130.0</generator><language>zh</language><copyright>©2024 zo0043</copyright><lastBuildDate>Wed, 30 Sep 2026 16:42:00 +0800</lastBuildDate><atom:link href="http://blog.zero43.top/tags/kv%E7%BC%93%E5%AD%98/index.xml" rel="self" type="application/rss+xml"/><item><title>DeepSeek 论文课（九）：长文本为什么贵：KV 压缩的四代路线之争</title><link>http://blog.zero43.top/posts/deepseek-paper-course-09-kv-compression/</link><pubDate>Wed, 30 Sep 2026 16:42:00 +0800</pubDate><guid>http://blog.zero43.top/posts/deepseek-paper-course-09-kv-compression/</guid><description>长文本为什么贵：KV 压缩的四代路线之争 先给结论： 模型读长文章要花的钱有两笔——一笔随文章变长一份一份往上加（KV cache），一笔随文章变长</description></item></channel></rss>