<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>大模型学习交流平台 · 从原理到实战</title>
    <link>https://guohua0siqi.online/</link>
    <description>从原理到实战的大模型学习笔记：Transformer、提示工程、RAG、Agent 与微调，配有结构化的学习路线和精选资源。</description>
    <language>zh-CN</language>
    <lastBuildDate>Thu, 17 Sep 2026 10:03:52 GMT</lastBuildDate>
    <atom:link href="https://guohua0siqi.online/rss.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>本地跑大模型：Ollama 上手与硬件选型</title>
      <link>https://guohua0siqi.online/blog/local-llm-ollama</link>
      <guid isPermaLink="true">https://guohua0siqi.online/blog/local-llm-ollama</guid>
      <pubDate>Sun, 06 Sep 2026 00:00:00 GMT</pubDate>
      <description>一条命令把模型跑在自己电脑上：量化等级怎么选、显存需要多少、什么时候该从 Ollama 换到 vLLM。</description>
      <category>tools</category>
      <category>Ollama</category>
      <category>本地部署</category>
      <category>量化</category>
      <category>vLLM</category>
    </item>
    <item>
      <title>AI Agent 的本质：一个 while 循环加几个工具</title>
      <link>https://guohua0siqi.online/blog/agent-basics</link>
      <guid isPermaLink="true">https://guohua0siqi.online/blog/agent-basics</guid>
      <pubDate>Fri, 28 Aug 2026 00:00:00 GMT</pubDate>
      <description>抛开框架看本质——Agent 就是「模型决定调哪个工具、你执行、把结果喂回去」的循环。附完整可跑的 Python 实现和四个必踩的坑。</description>
      <category>agent</category>
      <category>Agent</category>
      <category>工具调用</category>
      <category>Python</category>
      <category>Function Calling</category>
    </item>
    <item>
      <title>LoRA 微调实战：什么时候该微调，以及 4 个关键超参怎么调</title>
      <link>https://guohua0siqi.online/blog/lora-finetune</link>
      <guid isPermaLink="true">https://guohua0siqi.online/blog/lora-finetune</guid>
      <pubDate>Fri, 14 Aug 2026 00:00:00 GMT</pubDate>
      <description>LoRA 的低秩分解到底省了什么、r 和 alpha 怎么配、QLoRA 能省多少显存，以及微调之前你应该先确认的三件事。</description>
      <category>finetune</category>
      <category>微调</category>
      <category>LoRA</category>
      <category>QLoRA</category>
      <category>PEFT</category>
    </item>
    <item>
      <title>大模型选型：别只看榜单，先想清楚这四个问题</title>
      <link>https://guohua0siqi.online/blog/llm-model-selection</link>
      <guid isPermaLink="true">https://guohua0siqi.online/blog/llm-model-selection</guid>
      <pubDate>Sun, 26 Jul 2026 00:00:00 GMT</pubDate>
      <description>能力、成本、上下文、可用性——一套可复用的选型框架，以及「只盯着排行榜选模型」为什么几乎总会翻车。</description>
      <category>llm</category>
      <category>模型选型</category>
      <category>成本</category>
      <category>工程决策</category>
    </item>
    <item>
      <title>RAG 检索增强生成：原理不复杂，难的是工程细节</title>
      <link>https://guohua0siqi.online/blog/rag-in-practice</link>
      <guid isPermaLink="true">https://guohua0siqi.online/blog/rag-in-practice</guid>
      <pubDate>Thu, 09 Jul 2026 00:00:00 GMT</pubDate>
      <description>RAG 的五个环节怎么串起来、分块策略怎么选、什么时候该加重排、以及那些「Demo 很好用、上线就翻车」的典型原因。</description>
      <category>rag</category>
      <category>RAG</category>
      <category>向量检索</category>
      <category>工程实践</category>
    </item>
    <item>
      <title>提示工程的核心技巧：把「玄学」变成可复用的工程方法</title>
      <link>https://guohua0siqi.online/blog/prompt-engineering</link>
      <guid isPermaLink="true">https://guohua0siqi.online/blog/prompt-engineering</guid>
      <pubDate>Thu, 04 Jun 2026 00:00:00 GMT</pubDate>
      <description>结构化提示、少样本示例、思维链、XML 标签分隔、输出格式约束——把提示词当成接口来设计，而不是当成咒语来试。</description>
      <category>prompt</category>
      <category>提示工程</category>
      <category>Prompt</category>
      <category>最佳实践</category>
    </item>
    <item>
      <title>Transformer 与注意力机制：把「注意力」这件事讲清楚</title>
      <link>https://guohua0siqi.online/blog/transformer-attention</link>
      <guid isPermaLink="true">https://guohua0siqi.online/blog/transformer-attention</guid>
      <pubDate>Mon, 18 May 2026 00:00:00 GMT</pubDate>
      <description>从 QKV 的直觉、缩放点积注意力公式，到多头注意力和位置编码，用一段不到 40 行的 PyTorch 代码把 Transformer 的核心部件实现出来。</description>
      <category>basics</category>
      <category>Transformer</category>
      <category>注意力机制</category>
      <category>PyTorch</category>
    </item>
  </channel>
</rss>
