订阅
把下面的订阅地址添加到你常用的阅读器,即可接收站点更新。
订阅链接
订阅内容
<?xml version="1.0" encoding="utf-8"?>
<feed xmlns="http://www.w3.org/2005/Atom">
<id>https://blog.qiyang.dev/</id>
<title>纸上微光</title>
<updated>2026-05-26T00:00:00.000Z</updated>
<generator>Next.js + feed</generator>
<author>
<name>Qiyang(启阳)</name>
<uri>https://qiyang.dev/</uri>
</author>
<link rel="alternate" href="https://blog.qiyang.dev/"/>
<link rel="self" href="https://blog.qiyang.dev/atom.xml"/>
<subtitle>这里安放一些写下来的东西:技术、读书、生活、忽然冒出的念头,和那些暂时没有名字的片刻。文章不拘题材,像纸页接住风,也接住日常。</subtitle>
<icon>https://blog.qiyang.dev/favicon.ico</icon>
<rights>© 2026 Qiyang(启阳)</rights>
<entry>
<title type="html"><![CDATA[LLM 注意力机制:从点积到多头因果注意力]]></title>
<id>https://blog.qiyang.dev/posts/coding-attention-mechanisms</id>
<link href="https://blog.qiyang.dev/posts/coding-attention-mechanisms"/>
<updated>2026-05-26T00:00:00.000Z</updated>
<summary type="html"><![CDATA[基于《Build a Large Language Model (From Scratch)》第 3 章,整理注意力机制、自注意力、QKV、因果 mask、dropout 和多头注意力的实现路线,并把原章节图示重写为 Mermaid。]]></summary>
<category label="《Build a Large Language Model》阅读笔记" term="《Build a Large Language Model》阅读笔记"/>
<published>2026-05-26T00:00:00.000Z</published>
</entry>
<entry>
<title type="html"><![CDATA[如何读技术书更有收获(译文)]]></title>
<id>https://blog.qiyang.dev/posts/reading-recommendations</id>
<link href="https://blog.qiyang.dev/posts/reading-recommendations"/>
<updated>2026-05-26T00:00:00.000Z</updated>
<summary type="html"><![CDATA[译自 LLMs-from-scratch 仓库中关于技术书阅读方法的短文,整理按章节阅读、动手复现代码、做练习、回顾笔记,以及把想法用进项目的学习流程。]]></summary>
<category label="《Build a Large Language Model》阅读笔记" term="《Build a Large Language Model》阅读笔记"/>
<published>2026-05-26T00:00:00.000Z</published>
</entry>
<entry>
<title type="html"><![CDATA[LLM 读到的不是文字,而是一套序列协议]]></title>
<id>https://blog.qiyang.dev/posts/working-with-text-data</id>
<link href="https://blog.qiyang.dev/posts/working-with-text-data"/>
<updated>2026-05-26T00:00:00.000Z</updated>
<summary type="html"><![CDATA[从 tokenizer、token ID、特殊 token、滑动窗口采样、embedding、位置编码和 2026 年 API 协议出发,理解原始文本如何被改写成 LLM 可以计算的输入序列。]]></summary>
<category label="大语言模型" term="大语言模型"/>
<published>2026-05-24T00:00:00.000Z</published>
</entry>
<entry>
<title type="html"><![CDATA[LLM 定义、训练与 GPT 架构]]></title>
<id>https://blog.qiyang.dev/posts/understanding-large-language-models</id>
<link href="https://blog.qiyang.dev/posts/understanding-large-language-models"/>
<updated>2026-05-24T00:00:00.000Z</updated>
<summary type="html"><![CDATA[基于《Build a Large Language Model (From Scratch)》第 1 章,整理 LLM 的定义、应用、Transformer、GPT 架构、训练阶段和实现路线,并按 2026-05-24 的公开资料更新多模态、长上下文、后训练、MoE、开放权重和涌现能力等内容。]]></summary>
<category label="《Build a Large Language Model》阅读笔记" term="《Build a Large Language Model》阅读笔记"/>
<published>2026-05-22T00:00:00.000Z</published>
</entry>
<entry>
<title type="html"><![CDATA[CS146S 学习指南]]></title>
<id>https://blog.qiyang.dev/posts/cs146s-%E5%AD%A6%E4%B9%A0%E6%8C%87%E5%8D%97</id>
<link href="https://blog.qiyang.dev/posts/cs146s-%E5%AD%A6%E4%B9%A0%E6%8C%87%E5%8D%97"/>
<updated>2026-05-21T00:00:00.000Z</updated>
<summary type="html"><![CDATA[学习这门课的目标 理解 AI 的边界。 官方课程安排中的阅读资料有选择性的阅读即可。 第一周 Deep Dive into LLMs How OpenAI Uses Codex 第二周 阅读 MCP Introduction MCP Food-for-Thought 作业 我个人认为应该完成的作业是:Build a Custom MCP Server 第三周 Specs Are the New Source Code Writing Effective Tools for Agents]]></summary>
<category label="阅读笔记" term="阅读笔记"/>
<published>2026-05-21T00:00:00.000Z</published>
</entry>
</feed>