内容
<p align="center">
<img src="assets/banner.svg" alt="AI Engineering from Scratch · 简体中文版" width="100%">
</p>
<p align="center">
<b>从零开始，亲手实现每一个 AI 算法</b><br/>
<sub>523 节课 · 20 个阶段 · Python / TypeScript / Rust / Julia · 配套中文网站 <a href="https://aieng-zh.cn">aieng-zh.cn</a></sub>
</p>
<p align="center">
<a href="https://aieng-zh.cn"><img src="https://img.shields.io/badge/在线阅读-aieng--zh.cn-3553ff?style=flat-square&labelColor=fafaf5" alt="在线阅读 aieng-zh.cn"></a>
<a href="ROADMAP.md"><img src="https://img.shields.io/badge/lessons-523-3553ff?style=flat-square&labelColor=fafaf5" alt="523 lessons"></a>
<a href="#contents"><img src="https://img.shields.io/badge/phases-20-3553ff?style=flat-square&labelColor=fafaf5" alt="20 phases"></a>
<a href="LICENSE"><img src="https://img.shields.io/badge/license-MIT-1a1a1a?style=flat-square&labelColor=fafaf5" alt="MIT License"></a>
<a href="https://github.com/fancyboi999/ai-engineering-from-scratch-zh/stargazers"><img src="https://img.shields.io/github/stars/fancyboi999/ai-engineering-from-scratch-zh?style=flat-square&labelColor=fafaf5&color=3553ff&cacheSeconds=21600" alt="GitHub stars"></a>
</p>
```text
░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒
```
> 通过推导、代码和实验学习 AI 工程。
>
> 523 节课，20 个阶段，覆盖 Python、TypeScript、Rust、Julia。课程配有中文正文、测验、
> 可运行示例和可复用产物。免费阅读，开源，采用 MIT 许可。
>
> 从基础算法到 LLM 应用、agent 与生产系统，按你的目标选择学习路径。
> 本项目是 [AI Engineering from Scratch](https://github.com/rohitg00/ai-engineering-from-scratch)（作者 [Rohit Ghumare](https://github.com/rohitg00)，MIT 协议）的**简体中文衍生版**。衷心感谢原作者创作并开源了这套课程。
### 这个中文版做了什么
中文版提供课程翻译、配套视频、交互式学习网站和 AI 导师：
| | |
|---|---|
| 🇨🇳 **全站简体中文** | 523 节课正文、243 条术语表、测验题、`mermaid` 流程图、交互图表标签全部中文化（`agent`、`token`、`transformer` 等技术术语按惯例保留英文） |
| 🌐 **独立中文网站 [aieng-zh.cn](https://aieng-zh.cn)** | 可搜索的课程目录、[AI 工程学习路径](https://aieng-zh.cn/learning-paths.html)、学习进度追踪、可拖动的交互式图表、命令面板（`Cmd / Ctrl + K`）、深色模式 |
| 🎬 **配套动画讲解视频** | 333 节中文配音动画已上线，覆盖阶段 0–14 的全部课程。可在课程页内嵌播放，提供直链与 B 站播放源，配合正文、数学推导和代码实践学习 |
| 🔍 **为 AI 检索优化** | 构建时自动生成 `sitemap.xml` / `llms.txt` / 结构化数据，方便被搜索引擎和 AI 助手引用 |
| ✅ **课数一致性护栏** | CI 自动校验课程数（`node site/build.js --check`），防止课程列表与磁盘上的实际内容漂移 |
> 翻译怎么翻见 [TRANSLATION.md](TRANSLATION.md)。课程结构、代码与上游保持一致，译文持续跟进上游更新。
**目录** · [从这里开始](#从这里开始选择你想构建的内容) · [AI 导师](#30-秒添加-ai-导师) · [怎么运作](#怎么运作) · [课程结构](#课程的结构) · [学习路径](https://aieng-zh.cn/learning-paths.html) · [一节课的样子](#一节课的样子) · [快速开始](#快速开始) · [每节课都有产出](#每节课都有产出) · [课程目录](#contents) · [工具箱](#工具箱) · [参与贡献](#参与贡献)
## 从这里开始：选择你想构建的内容
开始前不需要先扫完 523 节课。选一个目标即可。每个链接都对应 GitHub 或网站上的同一套课程，两处使用相同的课程代码。
| 你的目标 | 在 GitHub 上学习 | 在网站上学习 |
|---|---|---|
| 我是新手，想要完整地打牢基础 | [阶段 0 · 配置与工具链](phases/00-setup-and-tooling/) | [开发环境](https://aieng-zh.cn/lessons/00-setup-and-tooling/01-dev-environment/) |
| 我懂 Python，想学数学与机器学习基础 | [阶段 1 · 数学基础](phases/01-math-foundations/) | [线性代数直觉](https://aieng-zh.cn/lessons/01-math-foundations/01-linear-algebra-intuition/) |
| 我想构建生产级 LLM 应用 | [阶段 11 · LLM 工程](phases/11-llm-engineering/) | [提示词工程](https://aieng-zh.cn/lessons/11-llm-engineering/01-prompt-engineering/) |
| 我想构建自主 agent | [阶段 14 · Agent 工程](phases/14-agent-engineering/) | [Agent 循环](https://aieng-zh.cn/lessons/14-agent-engineering/01-the-agent-loop/) |
| 我想在真实代码仓库中使用 coding agent | [Agent 辅助工程路径](learning-paths/using-coding-agents.json) | [Agent 辅助工程](https://aieng-zh.cn/lessons/14-agent-engineering/31-agent-workbench-why-models-fail/?learningPath=using-coding-agents) |
| 我想在编码实施前把产品方案想透定准 | [产品判断与交付路径](learning-paths/shaping-the-build.json) | [产品判断与交付](https://aieng-zh.cn/lessons/14-agent-engineering/47-outcomes-before-output/?learningPath=shaping-the-build) |
| 我想基于模型上下文协议（MCP）构建系统 | [模型上下文协议（MCP）路线](phases/13-tools-and-protocols/README.md#model-context-protocol-mcp-path) | [模型上下文协议（MCP）路径](https://aieng-zh.cn/lessons/13-tools-and-protocols/06-mcp-fundamentals/?learningPath=model-context-protocol) |
| 我想编写并交付 Agent Skills | [Agent Skills 快速路线](phases/13-tools-and-protocols/README.md#agent-skills-fast-path) | [Agent Skills 路径](https://aieng-zh.cn/lessons/13-tools-and-protocols/22-skills-and-agent-sdks/?learningPath=agent-skills) |
| 我想备考 Claude 官方认证 | [认证入门指南](certifications/claude/GETTING_STARTED.md) | [认证学院](https://aieng-zh.cn/certifications.html) |
| 我想备考 MCP Associate（MCPA） | [MCPA 入门指南](certifications/mcpa/GETTING_STARTED.md) | [MCPA 路线](https://aieng-zh.cn/certification?id=mcpa-f) |
拿不准自己适合从哪开始？使用 [`start-learning` 定位导师](skills/start-learning/SKILL.md) 或查看 [网站前置要求指南](https://aieng-zh.cn/prereqs.html)。
在 [AI 工程学习路径](https://aieng-zh.cn/learning-paths.html) 中对比四大核心领域与六大职业发展路线。
## 30 秒添加 AI 导师
```bash
npx skills add fancyboi999/ai-engineering-from-scratch-zh
```
| 宿主 | 开始全课程 | 开始模型上下文协议（MCP） | 开始 Agent Skills | 运行阶段测验 |
|---|---|---|---|---|
| Codex | `start-learning`，或从 `/skills` 选择 | `learn-mcp`，或从 `/skills` 选择 | `learn-agent-skills`，或从 `/skills` 选择 | `check-understanding 13`，或从 `/skills` 选择 |
| Claude Code | `/start-learning` | `/learn-mcp` | `/learn-agent-skills` | `/check-understanding 13` |
| 其他兼容宿主 | `使用 start-learning 开始课程。` | `使用 learn-mcp 开始模型上下文协议（MCP）路径。` | `使用 learn-agent-skills 开始 Agent Skills 工程路径。` | `使用 check-understanding 测验阶段 13。` |
运行你所在宿主的启动命令。导师会评估你当前的知识背景，把个性化学习计划保存到 `LEARNING.md`。从那里开始，`learn` 技能每次会话教学一节课：核心概念、数学直觉、编写代码、测验检验。它直接从本仓库流式读取课程，而 `course-guide` 技能可以在你卡住时精准跳转到对应课程。在 Codex 中，直接输入 `learn` 和 `course-guide`；在 Claude Code 中，输入 `/learn` 和 `/course-guide`；在其他兼容宿主中，按名称调用该技能即可。
只想要模型上下文协议（MCP）？使用宿主对应的 MCP 启动命令。它会生成 `MCP-LEARNING.md`，按照 [模型上下文协议（MCP）清单](learning-paths/model-context-protocol.json) 的 17 节课程路线，带你逐课实践无状态握手、服务端、客户端、传输层、采样、工具契约、可靠性与网关。
只想要 Agent Skills？使用宿主对应的 Agent Skills 启动命令。它会生成 `AGENT-SKILLS-LEARNING.md`，按照 [Agent Skills 路径](https://aieng-zh.cn/lessons/13-tools-and-protocols/22-skills-and-agent-sdks/?learningPath=agent-skills) 的 5 节必修课程路线，带你实践编写规范、发现机制、渐进式披露、安全沙箱与发布门禁。
| 技能 | 作用 |
|---|---|
| [`start-learning`](skills/start-learning/SKILL.md) | 定位导师。评估你当前的背景、设定学习目标、选择切入点，并生成 `LEARNING.md`。 |
| [`learn`](skills/learn/SKILL.md) | 动手教学导师。从仓库流式读取课程，带你推导概念、运行代码，并进行测验检验。 |
| [`learn-mcp`](skills/learn-mcp/SKILL.md) | MCP 专项导师。创建 `MCP-LEARNING.md`，遵循 17 课路线，记录协议线路、安全性、可靠性与一致性证据。 |
| [`learn-agent-skills`](skills/learn-agent-skills/SKILL.md) | Agent Skills 专项导师。创建 `AGENT-SKILLS-LEARNING.md`，遵循 5 课路线，记录技能契约、渐进式披露与真实 runner 测试证据。 |
| [`claude-certification`](skills/claude-certification/SKILL.md) | Claude 认证导师。选择 CCAO-F、CCDV-F、CCAR-F 或 CCAR-P，逐课教学、运行实验、评审产物、组织诊断与模拟测验并保存进度。 |
| [`mcpa-certification`](skills/mcpa-certification/SKILL.md) | MCPA 认证导师。遵循 34 课路线，运行 MCP 实验和协议校验，组织诊断与模拟测验，并保存学习进度。 |
| [`find-your-level`](skills/find-your-level/SKILL.md) | 十道题的定级测验。把你的知识映射到一个起始阶段，生成带课时估算的个性化路径。 |
| [`check-understanding <phase>`](skills/check-understanding/SKILL.md) | 按阶段测验，八道题，附反馈和需要复习的具体课程。 |
| [`course-guide`](skills/course-guide/SKILL.md) | 课程导引与概念查询。把任何主题或疑问映射到讲解该知识点的具体课程。 |
```text
░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒
```
## 怎么运作
大多数 AI 教材都是碎片化教学。这儿一篇论文，那儿一篇微调心得，别处再来个炫酷的 agent
demo。这些碎片很少能拼到一起。你做出了一个聊天机器人，却讲不清它的 loss 曲线；你给
agent 挂了个函数，却说不出调用它的那个模型内部，attention 到底在干什么。
这套课程就是那根脊椎。20 个阶段，523 节课，四种语言：Python、TypeScript、Rust、Julia。
一头是线性代数，另一头是自主 agent 集群。每个算法都先从最原始的数学手写出来。反向传播、
分词器、注意力、agent 循环——等 PyTorch 登场时，你已经知道它底层在做什么了。
学习时结合问题背景、概念推导、代码实验、测验和产出练习。动画帮助理解核心直觉，
代码与实验用于检验掌握程度。各课列出所需依赖；涉及 GPU、云服务或模型 API 的实践按课程要求准备资源，相关服务费用由提供商收取。
```text
░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒
```
## 课程的结构
二十个阶段层层叠起来。数学是地基，agent 和生产部署是屋顶。下层的东西你已经会了，就尽管
往前跳；但别跳过去之后，又回头纳闷上层为什么塌了。
```mermaid
%%{init: {'theme':'base','themeVariables':{'primaryColor':'#fafaf5','primaryTextColor':'#1a1a1a','primaryBorderColor':'#3553ff','lineColor':'#3553ff'}}}%%
flowchart TB
P0["阶段 0 · 配置与工具链"] --> P1["阶段 1 · 数学基础"]
P1 --> P2["阶段 2 · 机器学习基础"]
P2 --> P3["阶段 3 · 深度学习核心"]
P3 --> P4["阶段 4 · 计算机视觉"]
P3 --> P5["阶段 5 · NLP"]
P3 --> P6["阶段 6 · 语音与音频"]
P3 --> P9["阶段 9 · 强化学习"]
P5 --> P7["阶段 7 · Transformer"]
P7 --> P8["阶段 8 · 生成式 AI"]
P7 --> P10["阶段 10 · 从零实现 LLM"]
P10 --> P11["阶段 11 · LLM 工程"]
P10 --> P12["阶段 12 · 多模态 AI"]
P11 --> P13["阶段 13 · 工具与协议"]
P13 --> P14["阶段 14 · Agent 工程"]
P14 --> P15["阶段 15 · 自主系统"]
P15 --> P16["阶段 16 · 多 agent 与集群"]
P14 --> P17["阶段 17 · 基础设施与生产"]
P15 --> P18["阶段 18 · 伦理、安全与对齐"]
P16 --> P19["阶段 19 · 综合项目"]
P17 --> P19
P18 --> P19
```
```text
░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒
```
## 一节课的样子
每节课都待在自己的文件夹里，整套课程结构统一：
```text
phases/<NN>-<phase-name>/<NN>-<lesson-name>/
├── code/ 可运行的实现（Python、TypeScript、Rust、Julia）
├── docs/
│ └── zh.md 课程正文
├── quiz.json 课前、课中与课后测验
└── outputs/ 本节课的可复用产物
```
每节课都走六个节拍。其中 *Build It / Use It*（动手构建 / 实际使用）的拆分是整节课的脊椎——
你先从零实现算法，再用生产级的库把同样的事跑一遍。你之所以懂框架在做什么，是因为那个更小的
版本你自己写过。
```mermaid
%%{init: {'theme':'base','themeVariables':{'primaryColor':'#fafaf5','primaryTextColor':'#1a1a1a','primaryBorderColor':'#3553ff','lineColor':'#3553ff'}}}%%
flowchart LR
M["主旨<br/>一句话核心理念"] --> Pr["问题背景<br/>具体的痛点"]
Pr --> C["核心概念<br/>图解与直觉"]
C --> B["动手构建<br/>纯数学，不用框架"]
B --> U["实际使用<br/>同样的事用 PyTorch / sklearn 跑一遍"]
U --> S["拿去用<br/>提示词 · 技能 · agent · MCP"]
```
## 快速开始
三种入门方式。挑一个。
**方式 A —— 阅读。** 在
[aieng-zh.cn](https://aieng-zh.cn) 上打开任意一节已完成的课程，
或展开 [目录](#contents) 里的某个阶段。无需配置，无需 clone。
**方式 B —— clone 下来跑。**
```bash
git clone https://github.com/fancyboi999/ai-engineering-from-scratch-zh.git
cd ai-engineering-from-scratch-zh
python3 phases/00-setup-and-tooling/01-dev-environment/code/verify.py --route beginner
python3 phases/01-math-foundations/01-linear-algebra-intuition/code/vectors.py
```
**方式 C —— 让 AI 导师带你学。** 按上方[宿主调用表](#30-秒添加-ai-导师)调用技能。以下为 Claude Code 示例：
```bash
/start-learning # 评估背景、设定目标并生成个性化 LEARNING.md
/find-your-level # 十道题快速定级测验，映射起始阶段
```
按计划逐课学习，每学完一个阶段：
```bash
/check-understanding 3 # 测验你对阶段 3 的掌握
ls phases/03-deep-learning-core/05-loss-functions/outputs/
# ├── prompt-loss-function-selector.md
# └── prompt-loss-debugger.md
```
### 前置要求
- 你会写代码（任何语言都行，会 Python 更好）。
- 你想搞懂 AI **到底是怎么运作的**，而不只是调调 API。
### 准备 Claude 认证
[Claude 认证学院](certifications/claude/README.md) 是一套免费、开源的备考体系，
覆盖 Associate Foundations、Developer Foundations、Architect Foundations 和
Architect Professional 四条公开认证路线。每条路线都有对齐公开考试蓝图的课程、
可运行实验、诊断测验、综合项目和原创全真模拟题。
参照 [AI-native GitHub 入门指南](certifications/claude/GETTING_STARTED.md)，可以在 Claude Code、
Codex、ChatGPT、Cursor 或其他 agent 中调用 `claude-certification`（Claude Code 使用 `/claude-certification`）：选择路线、
把进度保存到 `CLAUDE-CERTIFICATION.md`，逐课学习、运行真实实验并接受基于产物的反馈。
同一套课程已在 [中文认证站点](https://aieng-zh.cn/certifications.html) 提供。
这是基于公开考试目标编写的独立学习材料，不隶属于 Anthropic，不包含真实考题，
也不保证通过认证。资格、费用、评分和项目政策可能变动，付费或预约前必须以官方最新说明为准。
### 准备 MCP Associate（MCPA）认证
[MCPA 认证课程](certifications/mcpa/README.md) 提供 34 节中文课程、可运行的 MCP 协议实验和原创练习题。课程依据 2026-07-28 版规范组织五个知识域；从 [入门指南](certifications/mcpa/GETTING_STARTED.md) 开始，使用 `mcpa-certification` 导师逐课练习，或在 [中文认证站点](https://aieng-zh.cn/certification?id=mcpa-f) 学习。
这套社区课程独立于 Agentic AI Foundation 和 Linux Foundation。正式考试政策以官方页面为准；模拟题成绩不等于正式考试分数。
### 内置 agent 技能
全套课程内置了 9 个涵盖定位、教学、专项路径与测验的 AI 导师技能，详见上文 [30 秒添加 AI 导师](#30-秒添加-ai-导师)；所有技能定义均位于 [`skills/`](skills/) 目录下。
```text
░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒
```
## 每节课都有产出
别的课程结尾是一句 *"恭喜，你学会了 X。"* 这里每节课的结尾，是一件你能直接装上、
或粘进日常工作流的 **可复用工具**。
<table>
<tr>
<th align="left" width="25%"><img src="site/assets/figures/001-a-prompts.svg" width="96" height="96" alt="FIG_001.A prompts"/><br/><sub>FIG_001 · A</sub><br/><b>PROMPTS</b></th>
<th align="left" width="25%"><img src="site/assets/figures/001-b-skills.svg" width="96" height="96" alt="FIG_001.B skills"/><br/><sub>FIG_001 · B</sub><br/><b>SKILLS</b></th>
<th align="left" width="25%"><img src="site/assets/figures/001-c-agents.svg" width="96" height="96" alt="FIG_001.C agents"/><br/><sub>FIG_001 · C</sub><br/><b>AGENTS</b></th>
<th align="left" width="25%"><img src="site/assets/figures/001-d-mcp-servers.svg" width="96" height="96" alt="FIG_001.D MCP servers"/><br/><sub>FIG_001 · D</sub><br/><b>MCP SERVERS</b></th>
</tr>
<tr>
<td valign="top">粘进任意 AI 助手，在某个细分任务上获得专家级帮助。</td>
<td valign="top">放进 Claude、Cursor、Codex、OpenClaw、Hermes，或任何能读 <code>SKILL.md</code> 的 agent。</td>
<td valign="top">作为自主 worker 部署——那个循环你在阶段 14 自己写过。</td>
<td valign="top">接入任意兼容 MCP 的客户端。在阶段 13 里从头到尾构建。</td>
</tr>
</table>
> 用 `python3 scripts/install_skills.py <target>` 将课程产出技能安装到指定目录。
> 各课的 `outputs/` 提供参考产物；把自己的实现、测试与复盘整理成作品集。
### FIG_002 · 一个实例
阶段 14，第 1 课：[Agent 循环](phases/14-agent-engineering/01-the-agent-loop/)。
Python 示例使用标准库和脚本化的 `ToyLLM`，可以离线观察工具调用、回合预算与停止条件。
```bash
python3 phases/14-agent-engineering/01-the-agent-loop/code/main.py
```
示例调用计算器和键值存储，输出执行轨迹，最终得到含 15% 税费的总价 `138.0`。
阅读 [Python 实现](phases/14-agent-engineering/01-the-agent-loop/code/main.py)、
[TypeScript 实现](phases/14-agent-engineering/01-the-agent-loop/code/main.ts)，以及
[可复用的 agent-loop 技能](phases/14-agent-engineering/01-the-agent-loop/outputs/skill-agent-loop.md)。
```text
░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒
```
<a id="contents"></a>
## 课程目录
二十个阶段。点开任意阶段即可展开它的课程列表。
<a id="phase-0"></a>
### Phase 0: 配置与工具链 `12 lessons`
> 把环境准备好，迎接后面所有的内容。
| # | Lesson | Type | Lang |
|:---:|--------|:----:|------|
| 01 | [开发环境](phases/00-setup-and-tooling/01-dev-environment/) | Build | Python |
| 02 | [Git 与协作](phases/00-setup-and-tooling/02-git-and-collaboration/) | Learn | — |
| 03 | [GPU 配置与云端](phases/00-setup-and-tooling/03-gpu-setup-and-cloud/) | Build | Python |
| 04 | [API 与密钥](phases/00-setup-and-tooling/04-apis-and-keys/) | Build | Python |
| 05 | [Jupyter Notebook](phases/00-setup-and-tooling/05-jupyter-notebooks/) | Build | Python |
| 06 | [Python 环境管理](phases/00-setup-and-tooling/06-python-environments/) | Build | Shell |
| 07 | [面向 AI 的 Docker](phases/00-setup-and-tooling/07-docker-for-ai/) | Build | Docker |
| 08 | [编辑器配置](phases/00-setup-and-tooling/08-editor-setup/) | Build | — |
| 09 | [数据管理](phases/00-setup-and-tooling/09-data-management/) | Build | Python |
| 10 | [终端与 Shell](phases/00-setup-and-tooling/10-terminal-and-shell/) | Learn | — |
| 11 | [面向 AI 的 Linux](phases/00-setup-and-tooling/11-linux-for-ai/) | Learn | — |
| 12 | [调试与性能分析](phases/00-setup-and-tooling/12-debugging-and-profiling/) | Build | Python |
<details id="phase-1">
<summary><b>Phase 1 — 数学基础</b> <code>22 lessons</code> <em>每个 AI 算法背后的直觉，用代码讲清楚。</em></summary>
<br/>
| # | Lesson | Type | Lang |
|:---:|--------|:----:|------|
| 01 | [线性代数直觉](phases/01-math-foundations/01-linear-algebra-intuition/) | Learn | Python, Julia |
| 02 | [向量、矩阵与运算](phases/01-math-foundations/02-vectors-matrices-operations/) | Build | Python, Julia |
| 03 | [矩阵变换与特征值](phases/01-math-foundations/03-matrix-transformations/) | Build | Python, Julia |
| 04 | [机器学习里的微积分：导数与梯度](phases/01-math-foundations/04-calculus-for-ml/) | Learn | Python |
| 05 | [链式法则与自动微分](phases/01-math-foundations/05-chain-rule-and-autodiff/) | Build | Python |
| 06 | [概率与分布](phases/01-math-foundations/06-probability-and-distributions/) | Learn | Python |
| 07 | [贝叶斯定理与统计思维](phases/01-math-foundations/07-bayes-theorem/) | Build | Python |
| 08 | [优化：梯度下降家族](phases/01-math-foundations/08-optimization/) | Build | Python |
| 09 | [信息论：熵与 KL 散度](phases/01-math-foundations/09-information-theory/) | Learn | Python |
| 10 | [降维：PCA、t-SNE、UMAP](phases/01-math-foundations/10-dimensionality-reduction/) | Build | Python |
| 11 | [奇异值分解](phases/01-math-foundations/11-singular-value-decomposition/) | Build | Python, Julia |
| 12 | [张量运算](phases/01-math-foundations/12-tensor-operations/) | Build | Python |
| 13 | [数值稳定性](phases/01-math-foundations/13-numerical-stability/) | Build | Python |
| 14 | [范数与距离](phases/01-math-foundations/14-norms-and-distances/) | Build | Python |
| 15 | [机器学习里的统计学](phases/01-math-foundations/15-statistics-for-ml/) | Build | Python |
| 16 | [采样方法](phases/01-math-foundations/16-sampling-methods/) | Build | Python |
| 17 | [线性方程组](phases/01-math-foundations/17-linear-systems/) | Build | Python |
| 18 | [凸优化](phases/01-math-foundations/18-convex-optimization/) | Build | Python |
| 19 | [面向 AI 的复数](phases/01-math-foundations/19-complex-numbers/) | Learn | Python |
| 20 | [傅里叶变换](phases/01-math-foundations/20-fourier-transform/) | Build | Python |
| 21 | [机器学习里的图论](phases/01-math-foundations/21-graph-theory/) | Build | Python |
| 22 | [随机过程](phases/01-math-foundations/22-stochastic-processes/) | Learn | Python |
</details>
<details id="phase-2">
<summary><b>Phase 2 — 机器学习基础</b> <code>18 lessons</code> <em>经典机器学习——至今仍是大多数生产 AI 的骨架。</em></summary>
<br/>
| # | Lesson | Type | Lang |
|:---:|--------|:----:|------|
| 01 | [什么是机器学习](phases/02-ml-fundamentals/01-what-is-machine-learning/) | Learn | Python |
| 02 | [从零实现线性回归](phases/02-ml-fundamentals/02-linear-regression/) | Build | Python |
| 03 | [逻辑回归与分类](phases/02-ml-fundamentals/03-logistic-regression/) | Build | Python |
| 04 | [决策树与随机森林](phases/02-ml-fundamentals/04-decision-trees/) | Build | Python |
| 05 | [支持向量机](phases/02-ml-fundamentals/05-support-vector-machines/) | Build | Python |
| 06 | [KNN 与距离度量](phases/02-ml-fundamentals/06-knn-and-distances/) | Build | Python |
| 07 | [无监督学习：K-Means、DBSCAN](phases/02-ml-fundamentals/07-unsupervised-learning/) | Build | Python |
| 08 | [特征工程与特征选择](phases/02-ml-fundamentals/08-feature-engineering/) | Build | Python |
| 09 | [模型评估：指标与交叉验证](phases/02-ml-fundamentals/09-model-evaluation/) | Build | Python |
| 10 | [偏差、方差与学习曲线](phases/02-ml-fundamentals/10-bias-variance/) | Learn | Python |
| 11 | [集成方法：Boosting、Bagging、Stacking](phases/02-ml-fundamentals/11-ensemble-methods/) | Build | Python |
| 12 | [超参数调优](phases/02-ml-fundamentals/12-hyperparameter-tuning/) | Build | Python |
| 13 | [机器学习流水线与实验追踪](phases/02-ml-fundamentals/13-ml-pipelines/) | Build | Python |
| 14 | [朴素贝叶斯](phases/02-ml-fundamentals/14-naive-bayes/) | Build | Python |
| 15 | [时间序列基础](phases/02-ml-fundamentals/15-time-series/) | Build | Python |
| 16 | [异常检测](phases/02-ml-fundamentals/16-anomaly-detection/) | Build | Python |
| 17 | [处理不平衡数据](phases/02-ml-fundamentals/17-imbalanced-data/) | Build | Python |
| 18 | [特征选择](phases/02-ml-fundamentals/18-feature-selection/) | Build | Python |
</details>
<details id="phase-3">
<summary><b>Phase 3 — 深度学习核心</b> <code>13 lessons</code> <em>从第一性原理出发的神经网络。先自己造一个，再碰框架。</em></summary>
<br/>
| # | Lesson | Type | Lang |
|:---:|--------|:----:|------|
| 01 | [感知机：一切的起点](phases/03-deep-learning-core/01-the-perceptron/) | Build | Python |
| 02 | [多层网络与前向传播](phases/03-deep-learning-core/02-multi-layer-networks/) | Build | Python |
| 03 | [从零实现反向传播](phases/03-deep-learning-core/03-backpropagation/) | Build | Python |
| 04 | [激活函数：ReLU、Sigmoid、GELU 及其原因](phases/03-deep-learning-core/04-activation-functions/) | Build | Python |
| 05 | [损失函数：MSE、交叉熵、对比损失](phases/03-deep-learning-core/05-loss-functions/) | Build | Python |
| 06 | [优化器：SGD、Momentum、Adam、AdamW](phases/03-deep-learning-core/06-optimizers/) | Build | Python |
| 07 | [正则化：Dropout、权重衰减、BatchNorm](phases/03-deep-learning-core/07-regularization/) | Build | Python |
| 08 | [权重初始化与训练稳定性](phases/03-deep-learning-core/08-weight-initialization/) | Build | Python |
| 09 | [学习率调度与 Warmup](phases/03-deep-learning-core/09-learning-rate-schedules/) | Build | Python |
| 10 | [造一个你自己的迷你框架](phases/03-deep-learning-core/10-mini-framework/) | Build | Python |
| 11 | [PyTorch 入门](phases/03-deep-learning-core/11-intro-to-pytorch/) | Build | Python |
| 12 | [JAX 入门](phases/03-deep-learning-core/12-intro-to-jax/) | Build | Python |
| 13 | [调试神经网络](phases/03-deep-learning-core/13-debugging-neural-networks/) | Build | Python |
</details>
<details id="phase-4">
<summary><b>Phase 4 — 计算机视觉</b> <code>28 lessons</code> <em>从像素到理解——图像、视频、3D、VLM 和世界模型。</em></summary>
<br/>
| # | Lesson | Type | Lang |
|:---:|--------|:----:|------|
| 01 | [图像基础：像素、通道、色彩空间](phases/04-computer-vision/01-image-fundamentals/) | Learn | Python |
| 02 | [从零实现卷积](phases/04-computer-vision/02-convolutions-from-scratch/) | Build | Python |
| 03 | [CNN：从 LeNet 到 ResNet](phases/04-computer-vision/03-cnns-lenet-to-resnet/) | Build | Python |
| 04 | [图像分类](phases/04-computer-vision/04-image-classification/) | Build | Python |
| 05 | [迁移学习与微调](phases/04-computer-vision/05-transfer-learning/) | Build | Python |
| 06 | [目标检测——从零实现 YOLO](phases/04-computer-vision/06-object-detection-yolo/) | Build | Python |
| 07 | [语义分割——U-Net](phases/04-computer-vision/07-semantic-segmentation-unet/) | Build | Python |
| 08 | [实例分割——Mask R-CNN](phases/04-computer-vision/08-instance-segmentation-mask-rcnn/) | Build | Python |
| 09 | [图像生成——GAN](phases/04-computer-vision/09-image-generation-gans/) | Build | Python |
| 10 | [图像生成——扩散模型](phases/04-computer-vision/10-image-generation-diffusion/) | Build | Python |
| 11 | [Stable Diffusion——架构与微调](phases/04-computer-vision/11-stable-diffusion/) | Build | Python |
| 12 | [视频理解——时序建模](phases/04-computer-vision/12-video-understanding/) | Build | Python |
| 13 | [3D 视觉：点云、NeRF](phases/04-computer-vision/13-3d-vision-nerf/) | Build | Python |
| 14 | [Vision Transformer（ViT）](phases/04-computer-vision/14-vision-transformers/) | Build | Python |
| 15 | [实时视觉：边缘部署](phases/04-computer-vision/15-real-time-edge/) | Build | Python |
| 16 | [构建一条完整的视觉流水线](phases/04-computer-vision/16-vision-pipeline-capstone/) | Build | Python |
| 17 | [自监督视觉——SimCLR、DINO、MAE](phases/04-computer-vision/17-self-supervised-vision/) | Build | Python |
| 18 | [开放词表视觉——CLIP](phases/04-computer-vision/18-open-vocab-clip/) | Build | Python |
| 19 | [OCR 与文档理解](phases/04-computer-vision/19-ocr-document-understanding/) | Build | Python |
| 20 | [图像检索与度量学习](phases/04-computer-vision/20-image-retrieval-metric/) | Build | Python |
| 21 | [关键点检测与姿态估计](phases/04-computer-vision/21-keypoint-pose/) | Build | Python |
| 22 | [从零实现 3D 高斯泼溅](phases/04-computer-vision/22-3d-gaussian-splatting/) | Build | Python |
| 23 | [Diffusion Transformer 与 Rectified Flow](phases/04-computer-vision/23-diffusion-transformers-rectified-flow/) | Build | Python |
| 24 | [SAM 3 与开放词表分割](phases/04-computer-vision/24-sam3-open-vocab-segmentation/) | Build | Python |
| 25 | [视觉语言模型（ViT-MLP-LLM）](phases/04-computer-vision/25-vision-language-models/) | Build | Python |
| 26 | [单目深度与几何估计](phases/04-computer-vision/26-monocular-depth/) | Build | Python |
| 27 | [多目标跟踪与视频记忆](phases/04-computer-vision/27-multi-object-tracking/) | Build | Python |
| 28 | [世界模型与视频扩散](phases/04-computer-vision/28-world-models-video-diffusion/) | Build | Python |
</details>
<details id="phase-5">
<summary><b>Phase 5 — NLP：从基础到进阶</b> <code>29 lessons</code> <em>语言是通往智能的接口。</em></summary>
<br/>
| # | Lesson | Type | Lang |
|:---:|--------|:----:|------|
| 01 | [文本处理：分词、词干提取、词形还原](phases/05-nlp-foundations-to-advanced/01-text-processing/) | Build | Python |
| 02 | [词袋、TF-IDF 与文本表示](phases/05-nlp-foundations-to-advanced/02-bag-of-words-tfidf/) | Build | Python |
| 03 | [词嵌入：从零实现 Word2Vec](phases/05-nlp-foundations-to-advanced/03-word-embeddings-word2vec/) | Build | Python |
| 04 | [GloVe、FastText 与子词嵌入](phases/05-nlp-foundations-to-advanced/04-glove-fasttext-subword/) | Build | Python |
| 05 | [情感分析](phases/05-nlp-foundations-to-advanced/05-sentiment-analysis/) | Build | Python |
| 06 | [命名实体识别（NER）](phases/05-nlp-foundations-to-advanced/06-named-entity-recognition/) | Build | Python |
| 07 | [词性标注与句法分析](phases/05-nlp-foundations-to-advanced/07-pos-tagging-parsing/) | Build | Python |
| 08 | [文本分类——用于文本的 CNN 与 RNN](phases/05-nlp-foundations-to-advanced/08-cnns-rnns-for-text/) | Build | Python |
| 09 | [序列到序列模型](phases/05-nlp-foundations-to-advanced/09-sequence-to-sequence/) | Build | Python |
| 10 | [注意力机制——那次突破](phases/05-nlp-foundations-to-advanced/10-attention-mechanism/) | Build | Python |
| 11 | [机器翻译](phases/05-nlp-foundations-to-advanced/11-machine-translation/) | Build | Python |
| 12 | [文本摘要](phases/05-nlp-foundations-to-advanced/12-text-summarization/) | Build | Python |
| 13 | [问答系统](phases/05-nlp-foundations-to-advanced/13-question-answering/) | Build | Python |
| 14 | [信息检索与搜索](phases/05-nlp-foundations-to-advanced/14-information-retrieval-search/) | Build | Python |
| 15 | [主题建模：LDA、BERTopic](phases/05-nlp-foundations-to-advanced/15-topic-modeling/) | Build | Python |
| 16 | [文本生成](phases/05-nlp-foundations-to-advanced/16-text-generation-pre-transformer/) | Build | Python |
| 17 | [聊天机器人：从规则到神经网络](phases/05-nlp-foundations-to-advanced/17-chatbots-rule-to-neural/) | Build | Python |
| 18 | [多语言 NLP](phases/05-nlp-foundations-to-advanced/18-multilingual-nlp/) | Build | Python |
| 19 | [子词分词：BPE、WordPiece、Unigram、SentencePiece](phases/05-nlp-foundations-to-advanced/19-subword-tokenization/) | Learn | Python |
| 20 | [结构化输出与约束解码](phases/05-nlp-foundations-to-advanced/20-structured-outputs-constrained-decoding/) | Build | Python |
| 21 | [自然语言推理与文本蕴含](phases/05-nlp-foundations-to-advanced/21-nli-textual-entailment/) | Learn | Python |
| 22 | [嵌入模型深入剖析](phases/05-nlp-foundations-to-advanced/22-embedding-models-deep-dive/) | Learn | Python |
| 23 | [RAG 的分块策略](phases/05-nlp-foundations-to-advanced/23-chunking-strategies-rag/) | Build | Python |
| 24 | [指代消解](phases/05-nlp-foundations-to-advanced/24-coreference-resolution/) | Learn | Python |
| 25 | [实体链接与消歧](phases/05-nlp-foundations-to-advanced/25-entity-linking/) | Build | Python |
| 26 | [关系抽取与知识图谱构建](phases/05-nlp-foundations-to-advanced/26-relation-extraction-kg/) | Build | Python |
| 27 | [LLM 评估：RAGAS、DeepEval、G-Eval](phases/05-nlp-foundations-to-advanced/27-llm-evaluation-frameworks/) | Build | Python |
| 28 | [长上下文评估：NIAH、RULER、LongBench、MRCR](phases/05-nlp-foundations-to-advanced/28-long-context-evaluation/) | Learn | Python |
| 29 | [对话状态跟踪](phases/05-nlp-foundations-to-advanced/29-dialogue-state-tracking/) | Build | Python |
</details>
<details id="phase-6">
<summary><b>Phase 6 — 语音与音频</b> <code>17 lessons</code> <em>听见、听懂、开口说。</em></summary>
<br/>
| # | Lesson | Type | Lang |
|:---:|--------|:----:|------|
| 01 | [音频基础：波形、采样、FFT](phases/06-speech-and-audio/01-audio-fundamentals) | Learn | Python |
| 02 | [频谱图、梅尔刻度与音频特征](phases/06-speech-and-audio/02-spectrograms-mel-features) | Build | Python |
| 03 | [音频分类](phases/06-speech-and-audio/03-audio-classification) | Build | Python |
| 04 | [语音识别（ASR）](phases/06-speech-and-audio/04-speech-recognition-asr) | Build | Python |
| 05 | [Whisper：架构与微调](phases/06-speech-and-audio/05-whisper-architecture-finetuning) | Build | Python |
| 06 | [说话人识别与验证](phases/06-speech-and-audio/06-speaker-recognition-verification) | Build | Python |
| 07 | [文本转语音（TTS）](phases/06-speech-and-audio/07-text-to-speech) | Build | Python |
| 08 | [声音克隆与音色转换](phases/06-speech-and-audio/08-voice-cloning-conversion) | Build | Python |
| 09 | [音乐生成](phases/06-speech-and-audio/09-music-generation) | Build | Python |
| 10 | [音频语言模型](phases/06-speech-and-audio/10-audio-language-models) | Build | Python |
| 11 | [实时音频处理](phases/06-speech-and-audio/11-real-time-audio-processing) | Build | Python |
| 12 | [搭一条语音助手流水线](phases/06-speech-and-audio/12-voice-assistant-pipeline) | Build | Python |
| 13 | [神经音频编解码器——EnCodec、SNAC、Mimi、DAC](phases/06-speech-and-audio/13-neural-audio-codecs) | Learn | Python |
| 14 | [语音活动检测与轮次切换](phases/06-speech-and-audio/14-voice-activity-detection-turn-taking) | Build | Python |
| 15 | [流式语音到语音——Moshi、Hibiki](phases/06-speech-and-audio/15-streaming-speech-to-speech-moshi-hibiki) | Learn | Python |
| 16 | [语音防伪与音频水印](phases/06-speech-and-audio/16-anti-spoofing-audio-watermarking) | Build | Python |
| 17 | [音频评估——WER、MOS、MMAU、排行榜](phases/06-speech-and-audio/17-audio-evaluation-metrics) | Learn | Python |
</details>
<details id="phase-7">
<summary><b>Phase 7 — Transformer 深入剖析</b> <code>16 lessons</code> <em>那个改变了一切的架构。</em></summary>
<br/>
| # | Lesson | Type | Lang |
|:---:|--------|:----:|------|
| 01 | [为什么用 Transformer：RNN 的问题](phases/07-transformers-deep-dive/01-why-transformers/) | Learn | Python |
| 02 | [从零实现自注意力](phases/07-transformers-deep-dive/02-self-attention-from-scratch/) | Build | Python |
| 03 | [多头注意力](phases/07-transformers-deep-dive/03-multi-head-attention/) | Build | Python |
| 04 | [位置编码：正弦、RoPE、ALiBi](phases/07-transformers-deep-dive/04-positional-encoding/) | Build | Python |
| 05 | [完整的 Transformer：编码器 + 解码器](phases/07-transformers-deep-dive/05-full-transformer/) | Build | Python |
| 06 | [BERT——掩码语言建模](phases/07-transformers-deep-dive/06-bert-masked-language-modeling/) | Build | Python |
| 07 | [GPT——因果语言建模](phases/07-transformers-deep-dive/07-gpt-causal-language-modeling/) | Build | Python |
| 08 | [T5、BART——编码器-解码器模型](phases/07-transformers-deep-dive/08-t5-bart-encoder-decoder/) | Learn | Python |
| 09 | [Vision Transformer（ViT）](phases/07-transformers-deep-dive/09-vision-transformers/) | Build | Python |
| 10 | [音频 Transformer——Whisper 架构](phases/07-transformers-deep-dive/10-audio-transformers-whisper/) | Learn | Python |
| 11 | [专家混合（MoE）](phases/07-transformers-deep-dive/11-mixture-of-experts/) | Build | Python |
| 12 | [KV Cache、Flash Attention 与推理优化](phases/07-transformers-deep-dive/12-kv-cache-flash-attention/) | Build | Python |
| 13 | [缩放定律](phases/07-transformers-deep-dive/13-scaling-laws/) | Learn | Python |
| 14 | [从零构建一个 Transformer](phases/07-transformers-deep-dive/14-build-a-transformer-capstone/) | Build | Python |
| 15 | [Attention 变体——滑动窗口、稀疏、差分](phases/07-transformers-deep-dive/15-attention-variants/) | Build | Python |
| 16 | [投机解码——草稿、验证、重复](phases/07-transformers-deep-dive/16-speculative-decoding/) | Build | Python |
</details>
<details id="phase-8">
<summary><b>Phase 8 — 生成式 AI</b> <code>15 lessons</code> <em>生成图像、视频、音频、3D，等等。</em></summary>
<br/>
| # | Lesson | Type | Lang |
|:---:|--------|:----:|------|
| 01 | [生成模型：分类与历史](phases/08-generative-ai/01-generative-models-taxonomy-history/) | Learn | Python |
| 02 | [自编码器与 VAE](phases/08-generative-ai/02-autoencoders-vae/) | Build | Python |
| 03 | [GAN：生成器 vs 判别器](phases/08-generative-ai/03-gans-generator-discriminator/) | Build | Python |
| 04 | [条件 GAN 与 Pix2Pix](phases/08-generative-ai/04-conditional-gans-pix2pix/) | Build | Python |
| 05 | [StyleGAN](phases/08-generative-ai/05-stylegan/) | Build | Python |
| 06 | [扩散模型——从零实现 DDPM](phases/08-generative-ai/06-diffusion-ddpm-from-scratch/) | Build | Python |
| 07 | [潜在扩散与 Stable Diffusion](phases/08-generative-ai/07-latent-diffusion-stable-diffusion/) | Build | Python |
| 08 | [ControlNet、LoRA 与条件控制](phases/08-generative-ai/08-controlnet-lora-conditioning/) | Build | Python |
| 09 | [图像修复、扩展与编辑](phases/08-generative-ai/09-inpainting-outpainting-editing/) | Build | Python |
| 10 | [视频生成](phases/08-generative-ai/10-video-generation/) | Build | Python |
| 11 | [音频生成](phases/08-generative-ai/11-audio-generation/) | Build | Python |
| 12 | [3D 生成](phases/08-generative-ai/12-3d-generation/) | Build | Python |
| 13 | [Flow Matching 与 Rectified Flow](phases/08-generative-ai/13-flow-matching-rectified-flows/) | Build | Python |
| 14 | [评估：FID、CLIP Score](phases/08-generative-ai/14-evaluation-fid-clip-score/) | Build | Python |
| 19 | [视觉自回归建模（VAR）：下一尺度预测](phases/08-generative-ai/19-visual-autoregressive-var/) | Build | Python |
</details>
<details id="phase-9">
<summary><b>Phase 9 — 强化学习</b> <code>12 lessons</code> <em>RLHF 和会玩游戏的 AI 的基石。</em></summary>
<br/>
| # | Lesson | Type | Lang |
|:---:|--------|:----:|------|
| 01 | [MDP、状态、动作与奖励](phases/09-reinforcement-learning/01-mdps-states-actions-rewards/) | Learn | Python |
| 02 | [动态规划](phases/09-reinforcement-learning/02-dynamic-programming/) | Build | Python |
| 03 | [蒙特卡洛方法](phases/09-reinforcement-learning/03-monte-carlo-methods/) | Build | Python |
| 04 | [Q-Learning、SARSA](phases/09-reinforcement-learning/04-q-learning-sarsa/) | Build | Python |
| 05 | [深度 Q 网络（DQN）](phases/09-reinforcement-learning/05-dqn/) | Build | Python |
| 06 | [策略梯度——REINFORCE](phases/09-reinforcement-learning/06-policy-gradients-reinforce/) | Build | Python |
| 07 | [Actor-Critic——A2C、A3C](phases/09-reinforcement-learning/07-actor-critic-a2c-a3c/) | Build | Python |
| 08 | [PPO](phases/09-reinforcement-learning/08-ppo/) | Build | Python |
| 09 | [奖励建模与 RLHF](phases/09-reinforcement-learning/09-reward-modeling-rlhf/) | Build | Python |
| 10 | [多智能体强化学习](phases/09-reinforcement-learning/10-multi-agent-rl/) | Build | Python |
| 11 | [仿真到现实的迁移](phases/09-reinforcement-learning/11-sim-to-real-transfer/) | Build | Python |
| 12 | [游戏中的强化学习](phases/09-reinforcement-learning/12-rl-for-games/) | Build | Python |
</details>
<details id="phase-10">
<summary><b>Phase 10 — 从零实现 LLM</b> <code>24 lessons</code> <em>构建、训练并真正理解大语言模型。</em></summary>
<br/>
| # | Lesson | Type | Lang |
|:---:|--------|:----:|------|
| 01 | [分词器：BPE、WordPiece、SentencePiece](phases/10-llms-from-scratch/01-tokenizers/) | Build | Python, Rust |
| 02 | [从零实现一个分词器](phases/10-llms-from-scratch/02-building-a-tokenizer/) | Build | Python |
| 03 | [预训练的数据流水线](phases/10-llms-from-scratch/03-data-pipelines/) | Build | Python |
| 04 | [预训练一个迷你 GPT（124M）](phases/10-llms-from-scratch/04-pre-training-mini-gpt/) | Build | Python |
| 05 | [分布式训练、FSDP、DeepSpeed](phases/10-llms-from-scratch/05-scaling-distributed/) | Build | Python |
| 06 | [指令微调——SFT](phases/10-llms-from-scratch/06-instruction-tuning-sft/) | Build | Python |
| 07 | [RLHF——奖励模型 + PPO](phases/10-llms-from-scratch/07-rlhf/) | Build | Python |
| 08 | [DPO——直接偏好优化](phases/10-llms-from-scratch/08-dpo/) | Build | Python |
| 09 | [Constitutional AI 与自我改进](phases/10-llms-from-scratch/09-constitutional-ai-self-improvement/) | Build | Python |
| 10 | [评估——基准与 evals](phases/10-llms-from-scratch/10-evaluation/) | Build | Python |
| 11 | [量化：INT8、GPTQ、AWQ、GGUF](phases/10-llms-from-scratch/11-quantization/) | Build | Python |
| 12 | [推理优化](phases/10-llms-from-scratch/12-inference-optimization/) | Build | Python |
| 13 | [搭一条完整的 LLM 流水线](phases/10-llms-from-scratch/13-building-complete-llm-pipeline/) | Build | Python |
| 14 | [开源模型：架构逐一拆解](phases/10-llms-from-scratch/14-open-models-architecture-walkthroughs/) | Learn | Python |
| 15 | [投机解码与 EAGLE-3](phases/10-llms-from-scratch/15-speculative-decoding-eagle3/) | Build | Python |
| 16 | [差分注意力（V2）](phases/10-llms-from-scratch/16-differential-attention-v2/) | Build | Python |
| 17 | [原生稀疏注意力（DeepSeek NSA）](phases/10-llms-from-scratch/17-native-sparse-attention/) | Build | Python |
| 18 | [多 token 预测（MTP）](phases/10-llms-from-scratch/18-multi-token-prediction/) | Build | Python |
| 19 | [DualPipe 并行](phases/10-llms-from-scratch/19-dualpipe-parallelism/) | Learn | Python |
| 20 | [DeepSeek-V3 架构拆解](phases/10-llms-from-scratch/20-deepseek-v3-walkthrough/) | Learn | Python |
| 21 | [Jamba——SSM-Transformer 混合架构](phases/10-llms-from-scratch/21-jamba-hybrid-ssm-transformer/) | Learn | Python |
| 22 | [异步与 Hogwild! 推理](phases/10-llms-from-scratch/22-async-hogwild-inference/) | Build | Python |
| 25 | [推测解码与 EAGLE](phases/10-llms-from-scratch/25-speculative-decoding/) | Build | Python |
| 34 | [梯度检查点与激活重算](phases/10-llms-from-scratch/34-gradient-checkpointing/) | Build | Python |
</details>
<details id="phase-11">
<summary><b>Phase 11 — LLM 工程</b> <code>17 lessons</code> <em>让 LLM 在生产环境里干活。</em></summary>
<br/>
| # | Lesson | Type | Lang |
|:---:|--------|:----:|------|
| 01 | [提示工程：技巧与套路](phases/11-llm-engineering/01-prompt-engineering/) | Build | Python |
| 02 | [Few-Shot、CoT、Tree-of-Thought](phases/11-llm-engineering/02-few-shot-cot/) | Build | Python |
| 03 | [结构化输出](phases/11-llm-engineering/03-structured-outputs/) | Build | Python |
| 04 | [嵌入与向量表示](phases/11-llm-engineering/04-embeddings/) | Build | Python |
| 05 | [上下文工程](phases/11-llm-engineering/05-context-engineering/) | Build | Python |
| 06 | [RAG：检索增强生成](phases/11-llm-engineering/06-rag/) | Build | Python |
| 07 | [进阶 RAG：分块、重排](phases/11-llm-engineering/07-advanced-rag/) | Build | Python |
| 08 | [用 LoRA 与 QLoRA 微调](phases/11-llm-engineering/08-fine-tuning-lora/) | Build | Python |
| 09 | [函数调用与工具使用](phases/11-llm-engineering/09-function-calling/) | Build | Python |
| 10 | [评估与测试](phases/11-llm-engineering/10-evaluation/) | Build | Python |
| 11 | [缓存、限流与成本](phases/11-llm-engineering/11-caching-cost/) | Build | Python |
| 12 | [护栏与安全](phases/11-llm-engineering/12-guardrails/) | Build | Python |
| 13 | [构建一个生产级 LLM 应用](phases/11-llm-engineering/13-production-app/) | Build | Python |
| 14 | [模型上下文协议（MCP）](phases/11-llm-engineering/14-model-context-protocol/) | Build | Python |
| 15 | [提示缓存与上下文缓存](phases/11-llm-engineering/15-prompt-caching/) | Build | Python |
| 16 | [Agent 状态机——图、节点与检查点](phases/11-llm-engineering/16-langgraph-state-machines/) | Build | Python |
| 17 | [agent 框架的取舍](phases/11-llm-engineering/17-agent-framework-tradeoffs/) | Learn | Python |
</details>
<details id="phase-12">
<summary><b>Phase 12 — 多模态 AI</b> <code>25 lessons</code> <em>跨模态地看、听、读、推理——从 ViT 的图块到操作电脑的 agent。</em></summary>
<br/>
| # | Lesson | Type | Lang |
|:---:|--------|:----:|------|
| 01 | [Vision Transformer 与图块-token 原语](phases/12-multimodal-ai/01-vision-transformer-patch-tokens/) | Learn | Python |
| 02 | [CLIP 与对比式视觉语言预训练](phases/12-multimodal-ai/02-clip-contrastive-pretraining/) | Build | Python |
| 03 | [BLIP-2 Q-Former 作为模态桥梁](phases/12-multimodal-ai/03-blip2-qformer-bridge/) | Build | Python |
| 04 | [Flamingo 与门控交叉注意力](phases/12-multimodal-ai/04-flamingo-gated-cross-attention/) | Learn | Python |
| 05 | [LLaVA 与视觉指令微调](phases/12-multimodal-ai/05-llava-visual-instruction-tuning/) | Build | Python |
| 06 | [任意分辨率视觉——Patch-n'-Pack 与 NaFlex](phases/12-multimodal-ai/06-any-resolution-patch-n-pack/) | Build | Python |
| 07 | [开源权重 VLM 配方：真正要紧的是什么](phases/12-multimodal-ai/07-open-weight-vlm-recipes/) | Learn | Python |
| 08 | [LLaVA-OneVision：单图、多图、视频](phases/12-multimodal-ai/08-llava-onevision-single-multi-video/) | Build | Python |
| 09 | [Qwen-VL 家族与动态 FPS 视频](phases/12-multimodal-ai/09-qwen-vl-family-dynamic-fps/) | Learn | Python |
| 10 | [InternVL3 原生多模态预训练](phases/12-multimodal-ai/10-internvl3-native-multimodal/) | Learn | Python |
| 11 | [Chameleon 早融合纯 token](phases/12-multimodal-ai/11-chameleon-early-fusion-tokens/) | Build | Python |
| 12 | [Emu3 用下一 token 预测做生成](phases/12-multimodal-ai/12-emu3-next-token-for-generation/) | Learn | Python |
| 13 | [Transfusion：自回归 + 扩散](phases/12-multimodal-ai/13-transfusion-autoregressive-diffusion/) | Build | Python |
| 14 | [Show-o 离散扩散统一架构](phases/12-multimodal-ai/14-show-o-discrete-diffusion-unified/) | Learn | Python |
| 15 | [Janus-Pro 解耦编码器](phases/12-multimodal-ai/15-janus-pro-decoupled-encoders/) | Build | Python |
| 16 | [MIO 任意到任意流式](phases/12-multimodal-ai/16-mio-any-to-any-streaming/) | Learn | Python |
| 17 | [视频语言时序定位](phases/12-multimodal-ai/17-video-language-temporal-grounding/) | Build | Python |
| 18 | [百万 token 上下文下的长视频](phases/12-multimodal-ai/18-long-video-million-token/) | Build | Python |
| 19 | [音频语言模型：从 Whisper 到 AF3](phases/12-multimodal-ai/19-audio-language-whisper-to-af3/) | Build | Python |
| 20 | [Omni 模型：Thinker-Talker 流式](phases/12-multimodal-ai/20-omni-models-thinker-talker/) | Build | Python |
| 21 | [具身 VLA：RT-2、OpenVLA、π0、GR00T](phases/12-multimodal-ai/21-embodied-vlas-openvla-pi0-groot/) | Learn | Python |
| 22 | [文档与图表理解](phases/12-multimodal-ai/22-document-diagram-understanding/) | Build | Python |
| 23 | [ColPali 视觉原生文档 RAG](phases/12-multimodal-ai/23-colpali-vision-native-rag/) | Build | Python |
| 24 | [多模态 RAG 与跨模态检索](phases/12-multimodal-ai/24-multimodal-rag-cross-modal/) | Build | Python |
| 25 | [多模态 agent 与操作电脑（综合项目）](phases/12-multimodal-ai/25-multimodal-agents-computer-use/) | Build | Python |
</details>
<details id="phase-13">
<summary><b>Phase 13 — 工具与协议</b> <code>31 lessons</code> <em>AI 与真实世界之间的接口，包括 Agent Skills 与无状态 MCP 的生产契约。</em></summary>
<br/>
| # | Lesson | Type | Lang |
|:---:|--------|:----:|------|
| 01 | [工具接口](phases/13-tools-and-protocols/01-the-tool-interface/) | Learn | Python |
| 02 | [函数调用深入剖析](phases/13-tools-and-protocols/02-function-calling-deep-dive/) | Build | Python |
| 03 | [并行与流式工具调用](phases/13-tools-and-protocols/03-parallel-and-streaming-tool-calls/) | Build | Python |
| 04 | [结构化输出](phases/13-tools-and-protocols/04-structured-output/) | Build | Python |
| 05 | [工具 Schema 设计](phases/13-tools-and-protocols/05-tool-schema-design/) | Learn | Python |
| 06 | [MCP 基础](phases/13-tools-and-protocols/06-mcp-fundamentals/) | Learn | Python |
| 07 | [构建一个 MCP server](phases/13-tools-and-protocols/07-building-an-mcp-server/) | Build | Python |
| 08 | [构建一个 MCP client](phases/13-tools-and-protocols/08-building-an-mcp-client/) | Build | Python |
| 09 | [MCP 传输层](phases/13-tools-and-protocols/09-mcp-transports/) | Learn | Python |
| 10 | [MCP 资源与提示](phases/13-tools-and-protocols/10-mcp-resources-and-prompts/) | Build | Python |
| 11 | [MCP Sampling](phases/13-tools-and-protocols/11-mcp-sampling/) | Build | Python |
| 12 | [MCP Roots 与 Elicitation](phases/13-tools-and-protocols/12-mcp-roots-and-elicitation/) | Build | Python |
| 13 | [MCP 异步任务](phases/13-tools-and-protocols/13-mcp-async-tasks/) | Build | Python |
| 14 | [MCP Apps](phases/13-tools-and-protocols/14-mcp-apps/) | Build | Python |
| 15 | [MCP 安全 I——工具投毒](phases/13-tools-and-protocols/15-mcp-security-tool-poisoning/) | Learn | Python |
| 16 | [MCP 安全 II——OAuth 2.1](phases/13-tools-and-protocols/16-mcp-security-oauth-2-1/) | Build | Python |
| 17 | [MCP 网关与注册表](phases/13-tools-and-protocols/17-mcp-gateways-and-registries/) | Learn | Python |
| 18 | [生产环境的 MCP 认证——iii 上的 DCR + JWKS](phases/13-tools-and-protocols/18-mcp-auth-production/) | Build | Python |
| 19 | [A2A 协议](phases/13-tools-and-protocols/19-a2a-protocol/) | Build | Python |
| 20 | [OpenTelemetry GenAI](phases/13-tools-and-protocols/20-opentelemetry-genai/) | Build | Python |
| 21 | [LLM 路由层](phases/13-tools-and-protocols/21-llm-routing-layer/) | Learn | Python |
| 22 | [Skills 与 Agent SDK](phases/13-tools-and-protocols/22-skills-and-agent-sdks/) | Learn | Python |
| 23 | [综合项目——工具生态](phases/13-tools-and-protocols/23-capstone-tool-ecosystem/) | Build | Python |
| 24 | [Skill 发现与渐进式披露](phases/13-tools-and-protocols/24-skill-discovery-and-progressive-disclosure/) | Build | Python |
| 25 | [Skill 调用与路由](phases/13-tools-and-protocols/25-skill-invocation-and-routing/) | Build | Python |
| 26 | [Skill 权限、沙箱与信任](phases/13-tools-and-protocols/26-skill-permissions-sandboxes-and-trust/) | Build | Python |
| 27 | [Skill 评测、打包与可移植性](phases/13-tools-and-protocols/27-skill-evals-packaging-and-portability/) | Build | Python |
| 28 | [MCP 工具契约与内容](phases/13-tools-and-protocols/28-mcp-tool-contracts-and-content/) | Build | Python |
| 29 | [MCP 可靠性、取消与流量控制](phases/13-tools-and-protocols/29-mcp-reliability-cancellation-and-flow-control/) | Build | Python |
| 30 | [MCP Registry 供应链：准入、漂移与回滚](phases/13-tools-and-protocols/30-mcp-registry-supply-chain-and-drift/) | Build | Python |
| 31 | [MCP 一致性工程：版本、证据与运维](phases/13-tools-and-protocols/31-mcp-conformance-versioning-and-operations/) | Build | Python |
[模型上下文协议（MCP）路径](learning-paths/model-context-protocol.json)包含第 06–18、28–31 课，共 17 节必修课，按清单顺序学习，其中第 18 课先于第 17 课。[Agent Skills 路径](learning-paths/agent-skills.json)包含第 22、24–27 课，共 5 节必修课。第 23 课是两条路径的可选系统综合项目，开始前按各路径清单完成前置要求。
</details>
<details id="phase-14">
<summary><b>Phase 14 — Agent 工程</b> <code>54 lessons</code> <em>从第一性原理构建 agent，可靠使用 coding agent，并在实现前把工作定义清楚。</em></summary>
<br/>
| # | Lesson | Type | Lang |
|:---:|--------|:----:|------|
| 01 | [Agent 循环](phases/14-agent-engineering/01-the-agent-loop/) | Build | Python |
| 02 | [ReWOO 与 Plan-and-Execute](phases/14-agent-engineering/02-rewoo-plan-and-execute/) | Build | Python |
| 03 | [Reflexion 与言语强化学习](phases/14-agent-engineering/03-reflexion-verbal-rl/) | Build | Python |
| 04 | [Tree of Thoughts 与 LATS](phases/14-agent-engineering/04-tree-of-thoughts-lats/) | Build | Python |
| 05 | [Self-Refine 与 CRITIC](phases/14-agent-engineering/05-self-refine-and-critic/) | Build | Python |
| 06 | [工具使用与函数调用](phases/14-agent-engineering/06-tool-use-and-function-calling/) | Build | Python |
| 07 | [Agent 记忆——虚拟上下文与记忆分页](phases/14-agent-engineering/07-memory-virtual-context-memgpt/) | Build | Python |
| 08 | [记忆块与休眠时计算](phases/14-agent-engineering/08-memory-blocks-sleep-time-compute/) | Build | Python |
| 09 | [混合记忆——向量 + 图 + KV](phases/14-agent-engineering/09-hybrid-memory-mem0/) | Build | Python |
| 10 | [技能库与终身学习——Voyager](phases/14-agent-engineering/10-skill-libraries-voyager/) | Build | Python |
| 11 | [用 HTN 与进化搜索做规划](phases/14-agent-engineering/11-planning-htn-and-evolutionary/) | Build | Python |
| 12 | [Anthropic 的工作流模式](phases/14-agent-engineering/12-anthropic-workflow-patterns/) | Build | Python |
| 13 | [有状态图编排——持久执行与检查点](phases/14-agent-engineering/13-langgraph-stateful-graphs/) | Build | Python |
| 14 | [Agent 的 Actor 模型——异步消息与类型化运行时](phases/14-agent-engineering/14-autogen-actor-model/) | Build | Python |
| 15 | [基于角色的 Agent 团队——角色、任务与流程](phases/14-agent-engineering/15-crewai-role-based-crews/) | Build | Python |
| 16 | [OpenAI Agents SDK——交接、护栏、追踪](phases/14-agent-engineering/16-openai-agents-sdk/) | Build | Python |
| 17 | [Harness 即库——子 agent 与会话存储](phases/14-agent-engineering/17-claude-agent-sdk/) | Build | Python |
| 18 | [生产级 Agent 运行时——快速实例化与类型化工作流](phases/14-agent-engineering/18-agno-and-mastra-runtimes/) | Learn | Python |
| 19 | [基准——SWE-bench、GAIA、AgentBench](phases/14-agent-engineering/19-benchmarks-swebench-gaia/) | Learn | Python |
| 20 | [基准——WebArena 与 OSWorld](phases/14-agent-engineering/20-benchmarks-webarena-osworld/) | Learn | Python |
| 21 | [操作电脑——Claude、OpenAI CUA、Gemini](phases/14-agent-engineering/21-computer-use-agents/) | Build | Python |
| 22 | [语音 agent——Pipecat 与 LiveKit](phases/14-agent-engineering/22-voice-agents-pipecat-livekit/) | Build | Python |
| 23 | [OpenTelemetry GenAI 语义约定](phases/14-agent-engineering/23-otel-genai-conventions/) | Build | Python |
| 24 | [Agent 可观测性——Langfuse、Phoenix、Opik](phases/14-agent-engineering/24-agent-observability-platforms/) | Learn | Python |
| 25 | [多 agent 辩论与协作](phases/14-agent-engineering/25-multi-agent-debate/) | Build | Python |
| 26 | [失败模式——agent 为什么会崩](phases/14-agent-engineering/26-failure-modes-agentic/) | Build | Python |
| 27 | [提示注入与 PVE 防御](phases/14-agent-engineering/27-prompt-injection-defense/) | Build | Python |
| 28 | [编排模式——Supervisor、Swarm、分层](phases/14-agent-engineering/28-orchestration-patterns/) | Build | Python |
| 29 | [生产级运行时——队列、事件、Cron](phases/14-agent-engineering/29-production-runtimes/) | Learn | Python |
| 30 | [Eval 驱动的 agent 开发](phases/14-agent-engineering/30-eval-driven-agent-development/) | Build | Python |
| 31 | [Agent 工作台：能力强的模型为什么仍会失败](phases/14-agent-engineering/31-agent-workbench-why-models-fail/) | Learn | Python |
| 32 | [最小化 agent 工作台](phases/14-agent-engineering/32-minimal-agent-workbench/) | Build | Python |
| 33 | [把 agent 指令写成可执行约束](phases/14-agent-engineering/33-instructions-as-executable-constraints/) | Build | Python |
| 34 | [仓库记忆与持久化状态](phases/14-agent-engineering/34-repo-memory-and-state/) | Build | Python |
| 35 | [给 agent 的初始化脚本](phases/14-agent-engineering/35-initialization-scripts/) | Build | Python |
| 36 | [范围契约与任务边界](phases/14-agent-engineering/36-scope-contracts/) | Build | Python |
| 37 | [运行时反馈回路](phases/14-agent-engineering/37-runtime-feedback-loops/) | Build | Python |
| 38 | [验证关卡](phases/14-agent-engineering/38-verification-gates/) | Build | Python |
| 39 | [审查 agent：把构建者和评判者分开](phases/14-agent-engineering/39-reviewer-agent/) | Build | Python |
| 40 | [多会话交接](phases/14-agent-engineering/40-multi-session-handoff/) | Build | Python |
| 41 | [在真实仓库上跑工作台](phases/14-agent-engineering/41-workbench-for-real-repos/) | Build | Python |
| 42 | [综合项目：交付一套可复用的 agent 工作台包](phases/14-agent-engineering/42-agent-workbench-capstone/) | Build | Python |
| 43 | [在 agent 写代码前框定任务](phases/14-agent-engineering/43-frame-the-task-before-code/) | Build | Python |
| 44 | [制定由证据支撑的执行计划](phases/14-agent-engineering/44-plan-from-evidence/) | Build | Python |
| 45 | [通过隔离与合并契约委派 agent 工作](phases/14-agent-engineering/45-delegate-with-isolation/) | Build | Python |
| 46 | [把每次 agent 纠正变成系统改进](phases/14-agent-engineering/46-turn-feedback-into-system/) | Build | Python |
| 47 | [在选择产出前定义结果](phases/14-agent-engineering/47-outcomes-before-output/) | Build | Python |
| 48 | [发现人们实际执行的工作流](phases/14-agent-engineering/48-discover-the-real-workflow/) | Build | Python |
| 49 | [绘制假设地图，先化解风险最高的一项](phases/14-agent-engineering/49-map-assumptions-and-risk/) | Build | Python |
| 50 | [选择能改变决定的最小可测试切片](phases/14-agent-engineering/50-choose-the-smallest-testable-slice/) | Build | Python |
| 51 | [编写保留判断空间的规格说明](phases/14-agent-engineering/51-write-specifications-that-preserve-judgment/) | Build | Python |
| 52 | [在结果出现前设计成功指标](phases/14-agent-engineering/52-design-success-metrics/) | Build | Python |
| 53 | [有意识地选择原型、试点或生产](phases/14-agent-engineering/53-prototype-pilot-or-production/) | Build | Python |
| 54 | [构建带所有权与淘汰机制的反馈棘轮](phases/14-agent-engineering/54-build-the-feedback-ratchet/) | Build | Python |
第 31-46 课组成 [Agent 辅助工程路径](learning-paths/using-coding-agents.json)。其清单顺序结合了工作台基础与任务框定、执行计划、隔离委派及长效反馈机制。第 47-54 课组成 [产品判断与交付路径](learning-paths/shaping-the-build.json)，涵盖从结果定义、事实证据、假设风险、切片选择、可执行规格说明到分阶段发布与反馈所有权的完整闭环。阶段 14 里每节工作台课程（31-42）都附带一份 `mission.md`，在 agent 打开完整课程文档前先给它做简报。
</details>
<details id="phase-15">
<summary><b>Phase 15 — 自主系统</b> <code>22 lessons</code> <em>长程 agent、自我改进，以及 2026 年的安全技术栈。</em></summary>
<br/>
| # | Lesson | Type | Lang |
|:---:|--------|:----:|------|
| 01 | [从聊天机器人到长程 agent（METR）](phases/15-autonomous-systems/01-long-horizon-agents/) | Learn | Python |
| 02 | [STaR、V-STaR、Quiet-STaR：自学推理](phases/15-autonomous-systems/02-star-family-reasoning/) | Learn | Python |
| 03 | [AlphaEvolve：进化式编码 agent](phases/15-autonomous-systems/03-alphaevolve-evolutionary-coding/) | Learn | Python |
| 04 | [Darwin Gödel Machine：自我修改的 agent](phases/15-autonomous-systems/04-darwin-godel-machine/) | Learn | Python |
| 05 | [AI Scientist v2：研讨会级别的科研](phases/15-autonomous-systems/05-ai-scientist-v2/) | Learn | Python |
| 06 | [自动化对齐研究（Anthropic AAR）](phases/15-autonomous-systems/06-automated-alignment-research/) | Learn | Python |
| 07 | [递归式自我改进：能力 vs 对齐](phases/15-autonomous-systems/07-recursive-self-improvement/) | Learn | Python |
| 08 | [有界自我改进的设计](phases/15-autonomous-systems/08-bounded-self-improvement/) | Learn | Python |
| 09 | [自主编码 agent 全景（SWE-bench、CodeAct）](phases/15-autonomous-systems/09-coding-agent-landscape/) | Learn | Python |
| 10 | [自主 Agent 的权限模式](phases/15-autonomous-systems/10-claude-code-permission-modes/) | Learn | Python |
| 11 | [浏览器 agent 与间接提示注入](phases/15-autonomous-systems/11-browser-agents/) | Learn | Python |
| 12 | [长时运行 agent 的持久化执行](phases/15-autonomous-systems/12-durable-execution/) | Learn | Python |
| 13 | [动作预算、迭代上限、成本管控](phases/15-autonomous-systems/13-cost-governors/) | Learn | Python |
| 14 | [急停开关、熔断器、金丝雀 token](phases/15-autonomous-systems/14-kill-switches-canaries/) | Learn | Python |
| 15 | [人在回路：先提议后提交](phases/15-autonomous-systems/15-propose-then-commit/) | Learn | Python |
| 16 | [检查点与回滚](phases/15-autonomous-systems/16-checkpoints-rollback/) | Learn | Python |
| 17 | [Constitutional AI 与规则覆盖](phases/15-autonomous-systems/17-constitutional-ai/) | Learn | Python |
| 18 | [Llama Guard 与输入/输出分类](phases/15-autonomous-systems/18-llama-guard/) | Learn | Python |
| 19 | [Anthropic 负责任扩展政策 v3.0](phases/15-autonomous-systems/19-anthropic-rsp/) | Learn | Python |
| 20 | [OpenAI Preparedness 框架与 DeepMind FSF](phases/15-autonomous-systems/20-openai-preparedness-deepmind-fsf/) | Learn | Python |
| 21 | [METR 时间跨度与外部评估](phases/15-autonomous-systems/21-metr-external-evaluation/) | Learn | Python |
| 22 | [CAIS、CAISI 与社会规模风险](phases/15-autonomous-systems/22-cais-caisi-societal-risk/) | Learn | Python |
</details>
<details id="phase-16">
<summary><b>Phase 16 — 多 agent 与集群</b> <code>25 lessons</code> <em>协调、涌现，以及集体智能。</em></summary>
<br/>
| # | Lesson | Type | Lang |
|:---:|--------|:----:|------|
| 01 | [为什么要多 agent](phases/16-multi-agent-and-swarms/01-why-multi-agent/) | Learn | TypeScript |
| 02 | [FIPA-ACL 传承与言语行为](phases/16-multi-agent-and-swarms/02-fipa-acl-heritage/) | Learn | Python |
| 03 | [通信协议](phases/16-multi-agent-and-swarms/03-communication-protocols/) | Build | TypeScript |
| 04 | [多 agent 原语模型](phases/16-multi-agent-and-swarms/04-primitive-model/) | Learn | Python |
| 05 | [Supervisor / 编排者-worker 模式](phases/16-multi-agent-and-swarms/05-supervisor-orchestrator-pattern/) | Build | Python |
| 06 | [分层架构与分解漂移](phases/16-multi-agent-and-swarms/06-hierarchical-architecture/) | Learn | Python |
| 07 | [心智社会与多 agent 辩论](phases/16-multi-agent-and-swarms/07-society-of-mind-debate/) | Build | Python |
| 08 | [角色专精——规划者 / 批评者 / 执行者 / 验证者](phases/16-multi-agent-and-swarms/08-role-specialization/) | Build | Python |
| 09 | [并行集群与网络化架构](phases/16-multi-agent-and-swarms/09-parallel-swarm-networks/) | Build | Python |
| 10 | [群聊与发言人选择](phases/16-multi-agent-and-swarms/10-group-chat-speaker-selection/) | Build | Python |
| 11 | [交接与例程（无状态编排）](phases/16-multi-agent-and-swarms/11-handoffs-and-routines/) | Build | Python |
| 12 | [A2A——Agent 到 Agent 协议](phases/16-multi-agent-and-swarms/12-a2a-protocol/) | Build | Python |
| 13 | [共享记忆与黑板模式](phases/16-multi-agent-and-swarms/13-shared-memory-blackboard/) | Build | Python |
| 14 | [共识与拜占庭容错](phases/16-multi-agent-and-swarms/14-consensus-and-bft/) | Build | Python |
| 15 | [投票、自洽性与辩论拓扑](phases/16-multi-agent-and-swarms/15-voting-debate-topology/) | Build | Python |
| 16 | [协商与议价](phases/16-multi-agent-and-swarms/16-negotiation-bargaining/) | Build | Python |
| 17 | [生成式 agent 与涌现式仿真](phases/16-multi-agent-and-swarms/17-generative-agents-simulation/) | Build | Python |
| 18 | [心智理论与涌现式协调](phases/16-multi-agent-and-swarms/18-theory-of-mind-coordination/) | Build | Python |
| 19 | [群体优化（PSO、ACO）](phases/16-multi-agent-and-swarms/19-swarm-optimization-pso-aco/) | Build | Python |
| 20 | [MARL——MADDPG、QMIX、MAPPO](phases/16-multi-agent-and-swarms/20-marl-maddpg-qmix-mappo/) | Learn | Python |
| 21 | [Agent 经济、token 激励、声誉](phases/16-multi-agent-and-swarms/21-agent-economies/) | Learn | Python |
| 22 | [生产级扩展——队列、检查点、持久性](phases/16-multi-agent-and-swarms/22-production-scaling-queues-checkpoints/) | Build | Python |
| 23 | [失败模式——MAST、群体思维、单一文化](phases/16-multi-agent-and-swarms/23-failure-modes-mast-groupthink/) | Learn | Python |
| 24 | [评估与协调基准](phases/16-multi-agent-and-swarms/24-evaluation-coordination-benchmarks/) | Learn | Python |
| 25 | [案例研究与 2026 最新进展](phases/16-multi-agent-and-swarms/25-case-studies-2026-sota/) | Learn | Python |
</details>
<details id="phase-17">
<summary><b>Phase 17 — 基础设施与生产</b> <code>28 lessons</code> <em>把 AI 交付到真实世界。</em></summary>
<br/>
| # | Lesson | Type | Lang |
|:---:|--------|:----:|------|
| 01 | [托管 LLM 平台 — Bedrock、Azure OpenAI、Vertex AI](phases/17-infrastructure-and-production/01-managed-llm-platforms/) | Learn | Python |
| 02 | [推理平台经济学 — Fireworks、Together、Baseten、Modal](phases/17-infrastructure-and-production/02-inference-platform-economics/) | Learn | Python |
| 03 | [Kubernetes 上的 GPU 自动扩缩 — Karpenter、KAI Scheduler](phases/17-infrastructure-and-production/03-gpu-autoscaling-kubernetes/) | Learn | Python |
| 04 | [推理服务引擎内部机制——PagedAttention、连续批处理与分块预填充](phases/17-infrastructure-and-production/04-vllm-serving-internals/) | Learn | Python |
| 05 | [生产环境中的 EAGLE-3 推测解码](phases/17-infrastructure-and-production/05-eagle3-speculative-decoding/) | Learn | Python |
| 06 | [前缀缓存推理服务——RadixAttention 与 KV 复用](phases/17-infrastructure-and-production/06-sglang-radixattention/) | Learn | Python |
| 07 | [硬件专用推理编译——Blackwell 上的 FP8 与 NVFP4](phases/17-infrastructure-and-production/07-tensorrt-llm-blackwell/) | Learn | Python |
| 08 | [推理指标 — TTFT、TPOT、ITL、Goodput、P99](phases/17-infrastructure-and-production/08-inference-metrics-goodput/) | Learn | Python |
| 09 | [生产级量化 — AWQ、GPTQ、GGUF、FP8、NVFP4](phases/17-infrastructure-and-production/09-production-quantization/) | Learn | Python |
| 10 | [无服务器 LLM 的冷启动缓解](phases/17-infrastructure-and-production/10-cold-start-mitigation/) | Learn | Python |
| 11 | [多区域 LLM 服务与 KV 缓存局部性](phases/17-infrastructure-and-production/11-multi-region-kv-locality/) | Learn | Python |
| 12 | [边缘推理 — ANE、Hexagon、WebGPU、Jetson](phases/17-infrastructure-and-production/12-edge-inference/) | Learn | Python |
| 13 | [LLM 可观测性技术栈选型](phases/17-infrastructure-and-production/13-llm-observability/) | Learn | Python |
| 14 | [提示缓存与语义缓存的经济学](phases/17-infrastructure-and-production/14-prompt-semantic-caching/) | Learn | Python |
| 15 | [批处理 API — 50% 折扣作为行业标准](phases/17-infrastructure-and-production/15-batch-apis/) | Learn | Python |
| 16 | [把模型路由作为降本原语](phases/17-infrastructure-and-production/16-model-routing/) | Learn | Python |
| 17 | [预填充/解码分离 — NVIDIA Dynamo 与 llm-d](phases/17-infrastructure-and-production/17-disaggregated-prefill-decode/) | Learn | Python |
| 18 | [生产级推理服务栈——KV 卸载与缓存感知路由](phases/17-infrastructure-and-production/18-vllm-production-stack-lmcache/) | Learn | Python |
| 19 | [AI 网关 — LiteLLM、Portkey、Kong、Bifrost](phases/17-infrastructure-and-production/19-ai-gateways/) | Learn | Python |
| 20 | [影子、金丝雀与渐进式部署](phases/17-infrastructure-and-production/20-shadow-canary-progressive/) | Learn | Python |
| 21 | [LLM 功能的 A/B 测试 — GrowthBook 与 Statsig](phases/17-infrastructure-and-production/21-ab-testing-llm-features/) | Learn | Python |
| 22 | [LLM API 的负载测试 — k6、LLMPerf、GenAI-Perf](phases/17-infrastructure-and-production/22-load-testing-llm-apis/) | Build | Python |
| 23 | [面向 AI 的 SRE — 多智能体事件响应](phases/17-infrastructure-and-production/23-sre-for-ai/) | Learn | Python |
| 24 | [面向 LLM 生产的混沌工程](phases/17-infrastructure-and-production/24-chaos-engineering-llm/) | Learn | Python |
| 25 | [安全 — 密钥、PII 脱敏、审计日志](phases/17-infrastructure-and-production/25-security-secrets-audit/) | Learn | Python |
| 26 | [合规 — SOC 2、HIPAA、GDPR、EU AI Act、ISO 42001](phases/17-infrastructure-and-production/26-compliance-frameworks/) | Learn | Python |
| 27 | [面向 LLM 的 FinOps — 单位经济与多租户归因](phases/17-infrastructure-and-production/27-finops-llms/) | Learn | Python |
| 28 | [自托管推理服务选型——让引擎匹配硬件与规模](phases/17-infrastructure-and-production/28-self-hosted-serving-selection/) | Learn | Python |
</details>
<details id="phase-18">
<summary><b>Phase 18 — 伦理、安全与对齐</b> <code>30 lessons</code> <em>构建对人类有益的 AI。这不是选修。</em></summary>
<br/>
| # | Lesson | Type | Lang |
|:---:|--------|:----:|------|
| 01 | [把遵循指令当作对齐信号](phases/18-ethics-safety-alignment/01-instruction-following-alignment-signal/) | Learn | Python |
| 02 | [奖励黑客与古德哈特定律](phases/18-ethics-safety-alignment/02-reward-hacking-goodhart/) | Learn | Python |
| 03 | [直接偏好优化家族](phases/18-ethics-safety-alignment/03-direct-preference-optimization-family/) | Learn | Python |
| 04 | [阿谀奉承：RLHF 的放大效应](phases/18-ethics-safety-alignment/04-sycophancy-rlhf-amplification/) | Learn | Python |
| 05 | [Constitutional AI 与 RLAIF](phases/18-ethics-safety-alignment/05-constitutional-ai-rlaif/) | Learn | Python |
| 06 | [Mesa 优化与欺骗性对齐](phases/18-ethics-safety-alignment/06-mesa-optimization-deceptive-alignment/) | Learn | Python |
| 07 | [潜伏 agent——持续性欺骗](phases/18-ethics-safety-alignment/07-sleeper-agents-persistent-deception/) | Learn | Python |
| 08 | [前沿模型中的上下文内谋划](phases/18-ethics-safety-alignment/08-in-context-scheming-frontier-models/) | Learn | Python |
| 09 | [对齐造假](phases/18-ethics-safety-alignment/09-alignment-faking/) | Learn | Python |
| 10 | [AI Control——即便被颠覆也保安全](phases/18-ethics-safety-alignment/10-ai-control-subversion/) | Learn | Python |
| 11 | [可扩展监督与弱到强](phases/18-ethics-safety-alignment/11-scalable-oversight-weak-to-strong/) | Learn | Python |
| 12 | [红队：PAIR 与自动化攻击](phases/18-ethics-safety-alignment/12-red-teaming-pair-automated-attacks/) | Build | Python |
| 13 | [多样本越狱](phases/18-ethics-safety-alignment/13-many-shot-jailbreaking/) | Learn | Python |
| 14 | [ASCII 字符画与视觉越狱](phases/18-ethics-safety-alignment/14-ascii-art-visual-jailbreaks/) | Build | Python |
| 15 | [间接提示注入](phases/18-ethics-safety-alignment/15-indirect-prompt-injection/) | Build | Python |
| 16 | [红队工具：Garak、Llama Guard、PyRIT](phases/18-ethics-safety-alignment/16-red-team-tooling-garak-llamaguard-pyrit/) | Build | Python |
| 17 | [WMDP 与双用途能力评估](phases/18-ethics-safety-alignment/17-wmdp-dual-use-evaluation/) | Learn | Python |
| 18 | [前沿安全框架——RSP、PF、FSF](phases/18-ethics-safety-alignment/18-frontier-safety-frameworks-rsp-pf-fsf/) | Learn | Python |
| 19 | [模型福祉研究](phases/18-ethics-safety-alignment/19-model-welfare-research/) | Learn | Python |
| 20 | [偏见与表征伤害](phases/18-ethics-safety-alignment/20-bias-representational-harm/) | Build | Python |
| 21 | [公平性准则：群体、个体、反事实](phases/18-ethics-safety-alignment/21-fairness-criteria-group-individual-counterfactual/) | Learn | Python |
| 22 | [面向 LLM 的差分隐私](phases/18-ethics-safety-alignment/22-differential-privacy-for-llms/) | Build | Python |
| 23 | [水印：SynthID、Stable Signature、C2PA](phases/18-ethics-safety-alignment/23-watermarking-synthid-stable-signature-c2pa/) | Build | Python |
| 24 | [监管框架：欧盟、美国、英国、韩国](phases/18-ethics-safety-alignment/24-regulatory-frameworks-eu-us-uk-korea/) | Learn | Python |
| 25 | [EchoLeak 与 AI 的 CVE](phases/18-ethics-safety-alignment/25-echoleak-cves-for-ai/) | Learn | Python |
| 26 | [模型卡、系统卡与数据集卡](phases/18-ethics-safety-alignment/26-model-system-dataset-cards/) | Build | Python |
| 27 | [数据溯源与训练数据治理](phases/18-ethics-safety-alignment/27-data-provenance-training-governance/) | Learn | Python |
| 28 | [对齐研究生态：MATS、Redwood、Apollo、METR](phases/18-ethics-safety-alignment/28-alignment-research-ecosystem/) | Learn | Python |
| 29 | [内容审核系统：OpenAI、Perspective、Llama Guard](phases/18-ethics-safety-alignment/29-moderation-systems-openai-perspective-llamaguard/) | Build | Python |
| 30 | [双用途风险：网络、生物、化学、核](phases/18-ethics-safety-alignment/30-dual-use-risk-cyber-bio-chem-nuclear/) | Learn | Python |
</details>
<details id="phase-19">
<summary><b>Phase 19 — 综合项目</b> <code>85 projects</code> <em>端到端项目与组件实践，按目标选择；各课正文列出预计时长。</em></summary>
<br/>
| # | Project | Combines | Lang |
|:---:|---------|----------|------|
| 01 | [终端原生编码 agent](phases/19-capstone-projects/01-terminal-native-coding-agent/) | P0 P5 P7 P10 P11 P13 P14 P15 P17 P18 | Python |
| 02 | [代码库 RAG（跨仓库语义搜索）](phases/19-capstone-projects/02-rag-over-codebase/) | P5 P7 P11 P13 P17 | Python |
| 03 | [实时语音助手（ASR → LLM → TTS）](phases/19-capstone-projects/03-realtime-voice-assistant/) | P6 P7 P11 P13 P14 P17 | Python |
| 04 | [多模态文档问答（视觉优先）](phases/19-capstone-projects/04-multimodal-document-qa/) | P4 P5 P7 P11 P12 P17 | Python |
| 05 | [自主科研 agent（AI-Scientist 级别）](phases/19-capstone-projects/05-autonomous-research-agent/) | P0 P2 P3 P7 P10 P14 P15 P16 P18 | Python |
| 06 | [面向 Kubernetes 的 DevOps 排障 agent](phases/19-capstone-projects/06-devops-troubleshooting-agent/) | P11 P13 P14 P15 P17 P18 | Python |
| 07 | [端到端微调流水线](phases/19-capstone-projects/07-end-to-end-fine-tuning-pipeline/) | P2 P3 P7 P10 P11 P17 P18 | Python |
| 08 | [生产级 RAG 聊天机器人（受监管垂直行业）](phases/19-capstone-projects/08-production-rag-chatbot/) | P5 P7 P11 P12 P17 P18 | Python |
| 09 | [代码迁移 agent（仓库级升级）](phases/19-capstone-projects/09-code-migration-agent/) | P5 P7 P11 P13 P14 P15 P17 | Python |
| 10 | [多 agent 软件工程团队](phases/19-capstone-projects/10-multi-agent-software-team/) | P11 P13 P14 P15 P16 P17 | Python |
| 11 | [LLM 可观测性与 Eval 仪表盘](phases/19-capstone-projects/11-llm-observability-dashboard/) | P11 P13 P17 P18 | Python |
| 12 | [视频理解流水线（场景 → 问答）](phases/19-capstone-projects/12-video-understanding-pipeline/) | P4 P6 P7 P11 P12 P17 | Python |
| 13 | [带注册表与治理的 MCP server](phases/19-capstone-projects/13-mcp-server-with-registry/) | P11 P13 P14 P17 P18 | Python |
| 14 | [投机解码推理服务器](phases/19-capstone-projects/14-speculative-decoding-server/) | P3 P7 P10 P17 | Python |
| 15 | [Constitutional 安全测试架 + 红队靶场](phases/19-capstone-projects/15-constitutional-safety-harness/) | P10 P11 P13 P14 P18 | Python |
| 16 | [GitHub Issue 到 PR 的自主 agent](phases/19-capstone-projects/16-github-issue-to-pr-agent/) | P11 P13 P14 P15 P17 | Python |
| 17 | [个人 AI 导师（自适应、多模态）](phases/19-capstone-projects/17-personal-ai-tutor/) | P5 P6 P11 P12 P14 P17 P18 | Python |
| 20 | [Agent Harness Loop 契约](phases/19-capstone-projects/20-agent-harness-loop-contract/) | A. Agent harness | Python |
| 21 | [带 Schema 校验的 Tool Registry](phases/19-capstone-projects/21-tool-registry-schema-validation/) | A. Agent harness | Python |
| 22 | [基于换行分隔 stdio 的 JSON-RPC 2.0](phases/19-capstone-projects/22-jsonrpc-stdio-transport/) | A. Agent harness | Python |
| 23 | [Function Call Dispatcher](phases/19-capstone-projects/23-function-call-dispatcher/) | A. Agent harness | Python |
| 24 | [Plan-Execute 控制流](phases/19-capstone-projects/24-plan-execute-control-flow/) | A. Agent harness | Python |
| 25 | [Verification Gate 与 Observation Budget](phases/19-capstone-projects/25-verification-gates-observation-budget/) | A. Agent harness | Python |
| 26 | [带 Denylist 与 Path Jail 的 Sandbox Runner](phases/19-capstone-projects/26-sandbox-runner-denylist/) | A. Agent harness | Python |
| 27 | [带 Fixture Tasks 的 Eval Harness](phases/19-capstone-projects/27-eval-harness-fixture-tasks/) | A. Agent harness | Python |
| 28 | [用 OTel GenAI Span 与 Prometheus 做 Observability](phases/19-capstone-projects/28-observability-otel-traces/) | A. Agent harness | Python |
| 29 | [端到端 Coding Agent Demo](phases/19-capstone-projects/29-end-to-end-coding-task-demo/) | A. Agent harness | Python |
| 30 | [从零实现 BPE Tokenizer](phases/19-capstone-projects/30-bpe-tokenizer-from-scratch/) | B. NLP LLM | Python |
| 31 | [带 Sliding Window 的 Tokenized Dataset](phases/19-capstone-projects/31-tokenized-dataset-sliding-window/) | B. NLP LLM | Python |
| 32 | [Token Embedding 与 Positional Embedding](phases/19-capstone-projects/32-token-positional-embeddings/) | B. NLP LLM | Python |
| 33 | [Multi-Head Self-Attention](phases/19-capstone-projects/33-multihead-self-attention/) | B. NLP LLM | Python |
| 34 | [从零实现 Transformer Block](phases/19-capstone-projects/34-transformer-block/) | B. NLP LLM | Python |
| 35 | [GPT 模型组装](phases/19-capstone-projects/35-gpt-model-assembly/) | B. NLP LLM | Python |
| 36 | [训练循环与评估](phases/19-capstone-projects/36-training-loop-eval/) | B. NLP LLM | Python |
| 37 | [加载预训练权重](phases/19-capstone-projects/37-loading-pretrained-weights/) | B. NLP LLM | Python |
| 38 | [通过换 Head 做分类微调](phases/19-capstone-projects/38-classifier-finetuning/) | B. NLP LLM | Python |
| 39 | [通过 SFT 做 Instruction Tuning](phases/19-capstone-projects/39-instruction-tuning-sft/) | B. NLP LLM | Python |
| 40 | [从零实现 DPO](phases/19-capstone-projects/40-dpo-from-scratch/) | B. NLP LLM | Python |
| 41 | [完整评估流水线](phases/19-capstone-projects/41-eval-pipeline/) | B. NLP LLM | Python |
| 42 | [大规模语料下载器](phases/19-capstone-projects/42-large-corpus-downloader/) | C. 端到端训练 | Python |
| 43 | [HDF5 Tokenized Corpus](phases/19-capstone-projects/43-hdf5-tokenized-corpus/) | C. 端到端训练 | Python |
| 44 | [Cosine 学习率 + 线性 Warmup](phases/19-capstone-projects/44-cosine-lr-warmup/) | C. 端到端训练 | Python |
| 45 | [Gradient Clipping 与混合精度训练](phases/19-capstone-projects/45-gradient-clipping-amp/) | C. 端到端训练 | Python |
| 46 | [梯度累积](phases/19-capstone-projects/46-gradient-accumulation/) | C. 端到端训练 | Python |
| 47 | [Checkpoint 保存与恢复](phases/19-capstone-projects/47-checkpoint-save-resume/) | C. 端到端训练 | Python |
| 48 | [从零实现分布式数据并行与 FSDP](phases/19-capstone-projects/48-distributed-fsdp-ddp/) | C. 端到端训练 | Python |
| 49 | [语言模型评测框架](phases/19-capstone-projects/49-lm-eval-harness/) | C. 端到端训练 | Python |
| 50 | [假设生成器](phases/19-capstone-projects/50-hypothesis-generator/) | D. 自动研究 | Python |
| 51 | [文献检索](phases/19-capstone-projects/51-literature-retrieval/) | D. 自动研究 | Python |
| 52 | [实验执行器](phases/19-capstone-projects/52-experiment-runner/) | D. 自动研究 | Python |
| 53 | [结果评估器](phases/19-capstone-projects/53-result-evaluator/) | D. 自动研究 | Python |
| 54 | [论文生成器](phases/19-capstone-projects/54-paper-writer/) | D. 自动研究 | Python |
| 55 | [评审循环](phases/19-capstone-projects/55-critic-loop/) | D. 自动研究 | Python |
| 56 | [迭代调度器](phases/19-capstone-projects/56-iteration-scheduler/) | D. 自动研究 | Python |
| 57 | [端到端研究 Demo](phases/19-capstone-projects/57-end-to-end-research-demo/) | D. 自动研究 | Python |
| 58 | [Vision Encoder 的 Patch 切分](phases/19-capstone-projects/58-vision-encoder-patches/) | E. 多模态 | Python |
| 59 | [Vision Transformer Encoder（ViT）](phases/19-capstone-projects/59-vit-transformer/) | E. 多模态 | Python |
| 60 | [用 Projection Layer 做模态对齐](phases/19-capstone-projects/60-projection-layer-modality-align/) | E. 多模态 | Python |
| 61 | [Cross-Attention 融合](phases/19-capstone-projects/61-cross-attention-fusion/) | E. 多模态 | Python |
| 62 | [Vision-Language 预训练](phases/19-capstone-projects/62-vision-language-pretraining/) | E. 多模态 | Python |
| 63 | [多模态评测](phases/19-capstone-projects/63-multimodal-eval/) | E. 多模态 | Python |
| 64 | [Chunking 策略横向对比](phases/19-capstone-projects/64-chunking-strategies-advanced/) | F. 高级 RAG | Python |
| 65 | [用 BM25 与 Dense Embedding 做 Hybrid Retrieval](phases/19-capstone-projects/65-hybrid-retrieval-bm25-dense/) | F. 高级 RAG | Python |
| 66 | [Cross-Encoder Reranker](phases/19-capstone-projects/66-reranker-cross-encoder/) | F. 高级 RAG | Python |
| 67 | [Query 改写：HyDE、Multi-Query 与 Decomposition](phases/19-capstone-projects/67-query-rewriting-hyde/) | F. 高级 RAG | Python |
| 68 | [RAG 评测：Precision、Recall、MRR、nDCG 等](phases/19-capstone-projects/68-rag-eval-precision-recall/) | F. 高级 RAG | Python |
| 69 | [端到端 RAG 系统](phases/19-capstone-projects/69-end-to-end-rag-system/) | F. 高级 RAG | Python |
| 70 | [任务规格格式](phases/19-capstone-projects/70-task-spec-format/) | G. 评测体系 | Python |
| 71 | [经典评测指标](phases/19-capstone-projects/71-classical-metrics/) | G. 评测体系 | Python |
| 72 | [代码执行评测指标](phases/19-capstone-projects/72-code-exec-metric/) | G. 评测体系 | Python |
| 73 | [perplexity 与 calibration](phases/19-capstone-projects/73-perplexity-calibration/) | G. 评测体系 | Python |
| 74 | [leaderboard 聚合](phases/19-capstone-projects/74-leaderboard-aggregation/) | G. 评测体系 | Python |
| 75 | [端到端 eval runner](phases/19-capstone-projects/75-end-to-end-eval-runner/) | G. 评测体系 | Python |
| 76 | [从零实现集合通信](phases/19-capstone-projects/76-collective-ops-from-scratch/) | H. 分布式训练 | Python |
| 77 | [数据并行 DDP](phases/19-capstone-projects/77-data-parallel-ddp/) | H. 分布式训练 | Python |
| 78 | [ZeRO Optimizer State 分片](phases/19-capstone-projects/78-zero-parameter-sharding/) | H. 分布式训练 | Python |
| 79 | [Pipeline Parallel 与 Bubble 分析](phases/19-capstone-projects/79-pipeline-parallel/) | H. 分布式训练 | Python |
| 80 | [分片 Checkpoint 与原子化恢复](phases/19-capstone-projects/80-checkpoint-sharded-resume/) | H. 分布式训练 | Python |
| 81 | [端到端分布式训练](phases/19-capstone-projects/81-end-to-end-distributed-train/) | H. 分布式训练 | Python |
| 82 | [越狱分类法](phases/19-capstone-projects/82-jailbreak-taxonomy/) | I. 安全护栏 | Python |
| 83 | [Prompt 注入检测器](phases/19-capstone-projects/83-prompt-injection-detector/) | I. 安全护栏 | Python |
| 84 | [拒答评估](phases/19-capstone-projects/84-refusal-evaluation/) | I. 安全护栏 | Python |
| 85 | [内容分类器集成](phases/19-capstone-projects/85-content-classifier-integration/) | I. 安全护栏 | Python |
| 86 | [Constitutional 规则引擎](phases/19-capstone-projects/86-constitutional-rules-engine/) | I. 安全护栏 | Python |
| 87 | [端到端 safety gate](phases/19-capstone-projects/87-end-to-end-safety-gate/) | I. 安全护栏 | Python |
</details>
```text
░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒
```
## 工具箱
课程产物保存在各课的 `phases/<phase>/<lesson>/outputs/` 下，包含提示词 Markdown、
技能 Markdown、完整 skill bundle 和项目参考文件。各课正文的“拿去用”章节说明产物用途；
下面分别说明导师技能与课程产出技能的安装方式。
### 把课程产出技能装进你的 agent
仓库在 `phases/**/outputs/` 下交付了 396 个技能和 99 个提示词。
**导师技能：通过 [skills.sh](https://skills.sh) 选择安装。** 默认发现 `skills/` 下的 9 个导师技能：
```bash
npx skills add fancyboi999/ai-engineering-from-scratch-zh --list
npx skills add fancyboi999/ai-engineering-from-scratch-zh --skill start-learning
```
**课程产出技能：克隆仓库后运行 `scripts/install_skills.py`。** 脚本读取课程的
`skill-*.md` 和完整 skill bundle，支持按阶段或标签筛选。将 `<target>` 替换为宿主的技能目录：
```bash
python3 scripts/install_skills.py <target> # 所有技能，默认 --layout skills（嵌套）
python3 scripts/install_skills.py <target> --layout skills # 同上，显式写出
python3 scripts/install_skills.py <target> --type all # 技能 + 提示词 + agent
python3 scripts/install_skills.py <target> --phase 14 # 只装一个阶段
python3 scripts/install_skills.py <target> --tag rag # 按标签过滤
python3 scripts/install_skills.py <target> --layout flat # 扁平文件
python3 scripts/install_skills.py <target> --dry-run # 只预览，不写入
python3 scripts/install_skills.py <target> --force # 覆盖已有文件
```
`<target>` 是你 agent 的技能目录（例如：
`~/.claude/skills/`、`~/.cursor/skills/`、`~/.config/openclaw/skills/`、
`.skills/`，或任何你 agent 读取的路径）。
默认情况下，脚本拒绝覆盖已存在的目标，会列出每个冲突路径后以退出码 1 退出。
用 `--dry-run` 预览冲突，或用 `--force` 覆盖。每次非 dry-run 的运行都会在目标里
写一份 `manifest.json`，按类型和阶段分组列出完整清单。挑你 agent 读取的那种布局：
| `--layout` | 写入路径 |
|---|---|
| `skills` | `<target>/<name>/SKILL.md`（嵌套约定，Claude / Cursor / Codex / OpenClaw / Hermes 都支持） |
| `by-phase` | `<target>/phase-NN/<name>.md` |
| `flat` | `<target>/<name>.md` |
### 把 agent 工作台塞进你自己的仓库
阶段 14 的综合项目交付了一套可复用的 Agent Workbench 包（AGENTS.md、schema、
init / verify / handoff 脚本）。用下面的命令把它脚手架进任意仓库：
```bash
python3 scripts/scaffold_workbench.py path/to/your-repo # 完整包 + 种子文件
python3 scripts/scaffold_workbench.py path/to/your-repo --minimal # 跳过 docs/
python3 scripts/scaffold_workbench.py path/to/your-repo --dry-run # 只预览
python3 scripts/scaffold_workbench.py path/to/your-repo --force # 覆盖
```
你会得到接好线的七个工作台界面、一份起步用的 `task_board.json`，
以及一份 `schema_version: 1` 的全新 `agent_state.json`。从这里开始：编辑任务、
编辑 `AGENTS.md`、运行 `scripts/init_agent.py`，把契约交给你的 agent。包的源码在
`phases/14-agent-engineering/42-agent-workbench-capstone/outputs/agent-workbench-pack/`。
### 课程数据与课数校验
`site/build.js` 解析 README、ROADMAP 和每节课的 `docs/zh.md`，在 `site/data.js`
里生成站点用的课程数据（阶段、课程、术语表、产物）。计数以**文件系统**为真相。
```bash
node site/build.js # 生成 site/data.js（+ sitemap.xml / llms.txt）
node site/build.js --check # 只校验课程数一致性，不写文件
```
`--check` 以磁盘上的课程目录数为准，核对 README 表格、badge、散文、各 phase 标题
和 ROADMAP 总计是否都对得上，任一漂移就 exit 1。一个 GitHub Action
（`.github/workflows/build.yml`）在每个 PR 上跑构建与课数校验。
新增课程时在 README + ROADMAP 表格补行、并更新该 phase
标题的课数即可；站点模板里散落的计数由 build 时 `syncCounts` 自动同步。
> 注：`scripts/build_catalog.py` 也已支持 `docs/zh.md`，但 `catalog.json` 只是可再生的辅助产物；
> 中文站发布与课程计数仍以 `site/build.js`、`scripts/audit_lessons.py` 和 README / ROADMAP 为准。
### 给每节课的 Python 代码做冒烟检查
`scripts/lesson_run.py` 会逐字节编译每节课 `code/` 目录下的每个 `.py` 文件。
默认模式只做语法检查——不执行、不需要 API key、不需要重型 ML 依赖。专抓贡献者最常
引入的回归（缩进错误、f-string 写坏、误改）。
```bash
python3 scripts/lesson_run.py # 语法检查整套课程
python3 scripts/lesson_run.py --phase 14 # 只查一个阶段
python3 scripts/lesson_run.py --json # 在 stdout 输出 JSON 报告
python3 scripts/lesson_run.py --strict # 任一课程失败就 exit 1
python3 scripts/lesson_run.py --execute # 真正运行，每节课 10 秒超时
```
`--execute` 会运行每节课的 `code/main.py`（或第一个 `.py` 文件），10 秒超时。
入口文件开头带 `# requires: pkg1, pkg2` 注释（列出非标准库依赖）的课程会被跳过，
原因标为 `needs <deps>`。这个脚本是可选的，没接入 CI。
仅用标准库，Python 3.10+。设置 `LINK_CHECK_SKIP=domain1,domain2` 可以覆盖默认跳过名单
（`twitter.com`、`x.com`、`linkedin.com`、`instagram.com`、`medium.com`——这些域名
会主动屏蔽自动化的 HEAD/GET）。
## 从哪里开始
| 你的背景 | 从哪开始 | 学习范围 |
|---|---|---|
| 编程和 AI 都是新手 | 阶段 0 — 配置与工具链 | 先准备环境，再按前置依赖学习 |
| 会 Python，刚接触 ML | 阶段 1 — 数学基础 | 数学与机器学习基础 |
| 懂 ML，刚接触深度学习 | 阶段 3 — 深度学习核心 | 神经网络与训练 |
| 懂深度学习，想学 LLM 和 agent | 阶段 10 — 从零实现 LLM | 模型、应用与 agent |
| 资深工程师，只想要 agent 工程 | 阶段 14 — Agent 工程 | 按目标选择工程课程与综合项目 |
| 只想构建生产级 MCP 系统 | [模型上下文协议（MCP）路径](learning-paths/model-context-protocol.json) | 17 节必修，约 23 小时 15 分 |
| 只想构建生产级 Agent Skills | [Agent Skills 工程路径](learning-paths/agent-skills.json) | 5 节必修，约 9 小时 30 分 |
[ROADMAP.md](ROADMAP.md) 按各课正文汇总预计时长。阶段 0–18 的 438 节课合计约 542 小时；
阶段 19 的 85 项综合项目与组件实践合计约 627 小时，适合按目标选做。
专项路径以 `learning-paths/*.json` 的必修清单和时长为准。实际投入取决于基础、实验环境和项目深度。
```text
░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒
```
## 为什么这件事现在很重要
<table>
<tr>
<th align="left" width="50%"><sub>FIG_003 · A</sub><br/><b>THE INDUSTRY SIGNAL</b></th>
<th align="left" width="50%"><sub>FIG_003 · B</sub><br/><b>FOUNDATIONAL PAPERS COVERED</b></th>
</tr>
<tr>
<td valign="top">
> *"最热门的新编程语言，是英语。"*<br/>
> — **Andrej Karpathy** ([tweet](https://x.com/karpathy/status/1617979122625712128))
课程围绕可运行的实现组织学习：阅读论文与规范，构建最小示例，再用测试和产物检查理解。
</td>
<td valign="top">
- *Attention Is All You Need* — Vaswani et al., 2017 → [Phase 7](#phase-7)
- *Language Models are Few-Shot Learners* (GPT-3) → [Phase 10](#phase-10)
- *Denoising Diffusion Probabilistic Models* → [Phase 8](#phase-8)
- *InstructGPT / RLHF* → [Phase 10](#phase-10)
- *Direct Preference Optimization* → [Phase 10](#phase-10)
- *Chain-of-Thought Prompting* → [Phase 11](#phase-11)
- *ReAct: Reasoning + Acting in LLMs* → [Phase 14](#phase-14)
- *Model Context Protocol* — Anthropic → [Phase 13](#phase-13)
</td>
</tr>
</table>
```text
░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒░░░▒▒▒
```
## 参与贡献
| Goal | Read |
|---|---|
| 贡献一节课或一处修复 | [CONTRIBUTING.md](CONTRIBUTING.md) |
| 为你的团队或学校 fork | [FORKING.md](FORKING.md) |
| 课程模板 | [LESSON_TEMPLATE.md](LESSON_TEMPLATE.md) |
| 跟踪进度 | [ROADMAP.md](ROADMAP.md) |
| 术语表 | [glossary/terms.md](glossary/terms.md) |
| 行为准则 | [CODE_OF_CONDUCT.md](CODE_OF_CONDUCT.md) |
提交一节课之前，先跑一遍不变量检查：
```bash
python3 scripts/audit_lessons.py # 整套课程
python3 scripts/audit_lessons.py --phase 14 # 单个阶段
python3 scripts/audit_lessons.py --json # 适合 CI 的输出
```
任一规则失败时退出码非零。规则（L001–L010）会校验目录结构、`docs/zh.md` 是否存在
及是否有 H1、`code/` 是否非空、`quiz.json` 的 schema，以及课程文档里的相对链接。
## Star 历史
<a href="https://star-history.com/#fancyboi999/ai-engineering-from-scratch-zh&Date">
<picture>
<source media="(prefers-color-scheme: dark)" srcset="https://api.star-history.com/svg?repos=fancyboi999/ai-engineering-from-scratch-zh&type=Date&theme=dark">
<img alt="Star history" src="https://api.star-history.com/svg?repos=fancyboi999/ai-engineering-from-scratch-zh&type=Date" width="100%">
</picture>
</a>
如果这份手册帮到了你，给仓库点个 star。这能让项目活下去。
## 许可
采用 [MIT 许可](LICENSE)，允许使用、修改和商业分发。分发副本或实质性部分时，须保留版权声明和许可声明。
原课程由 [Rohit Ghumare](https://github.com/rohitg00) 及社区创建，简体中文版由 [fancy](https://github.com/fancyboi999) 维护。
<sub>
<a href="https://aieng-zh.cn">aieng-zh.cn</a> ·
<a href="https://github.com/fancyboi999/ai-engineering-from-scratch-zh/issues/new/choose">Report / Suggest</a>
</sub>
连接信息
猜你喜欢
暂无推荐服务器