<?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>LLM 推理 on Hellokitty&#39;s Blog</title>
    <link>https://yangyang233333.github.io/tags/llm-%E6%8E%A8%E7%90%86/</link>
    <description>Recent content in LLM 推理 on Hellokitty&#39;s Blog</description>
    <generator>Hugo</generator>
    <language>zh-CN</language>
    <lastBuildDate>Sat, 12 Sep 2026 22:40:00 +0800</lastBuildDate>
    <atom:link href="https://yangyang233333.github.io/tags/llm-%E6%8E%A8%E7%90%86/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>从提示词到下一个 token：nano-vLLM 请求生命周期、KV cache 与 batch 时序</title>
      <link>https://yangyang233333.github.io/posts/nano-vllm-request-lifecycle/</link>
      <pubDate>Sat, 12 Sep 2026 22:40:00 +0800</pubDate>
      <guid>https://yangyang233333.github.io/posts/nano-vllm-request-lifecycle/</guid>
      <description>沿一次请求追踪 chat template、tokenizer、调度、GPU 输入打包、Transformer 计算、分页 KV cache、采样与返回，并用源码级实验厘清前缀命中和 batch overlap 的真实边界。</description>
    </item>
  </channel>
</rss>
