LLM Cookbook: Andrew Ng's LLM Courses in Chinese
/llm-cookbook
LLM Cookbook is a Chinese translation and reproduction of Andrew Ng's DeepLearning.AI short courses, worth the time if you read Chinese and want a graded path through prompt engineering, LangChain, and RAG instead of scattered blog posts. The curriculum sorts every course into required or elective. It stops short of model internals: this is API-and-framework practice, not a from-scratch ML course.
LLM Cookbook 是什么
LLM Cookbook is Datawhale's Chinese translation and reproduction of 4 required and 8 elective DeepLearning.AI short courses built with Andrew Ng, covering prompt engineering, ChatGPT app building, LangChain, RAG, Gradio interfaces, and fine-tuning. Each course ships as a runnable Jupyter notebook, and the required track also gets a PDF. It's for developers with basic Python who want a Chinese on-ramp into LLM API development.
11 门课程如何分级：必修与选修路径
- ✓Four required courses cover the baseline skills every LLM developer needs: prompt construction, a full ChatGPT-based Q&A system, and two LangChain courses on building apps and querying your own data.
- ✓Eight elective courses branch into Gradio UIs, evaluating generative AI with W&B, fine-tuning open models with the lamini framework, and four separate courses on retrieval: plain semantic search, Chroma-based retrieval, full RAG evaluation, and LangChain agents with Functions and Tools.
- ✓Every course is reproduced as runnable notebooks in the content/ folder, which the README says updates most often, while docs/ holds a Markdown reading version of just the required track.
- ✓The team re-ran their own comparison tests to write Chinese prompts the README says perform about as well as the English originals, instead of just machine-translating Andrew Ng's prompt text.
- ✓Only the Prompt Engineering course has a separately produced bilingual-subtitle video on Bilibili, plus a downloadable bilingual subtitle file from a companion repo; the other ten courses skip that.
- ✓There's a standalone PDF export of the required-course track, released as a tagged GitHub release, for reading offline instead of in a notebook.
学完能做什么：从 API 调用到生产级 RAG 应用
- •Writing structured prompts for summarizing, inferring sentiment, and transforming text via the OpenAI API, straight out of the required Prompt Engineering course.
- •Building a full customer-support-style Q&A system on top of the ChatGPT API, following the second required course end to end.
- •Standing up a RAG pipeline that answers questions from your own documents, combining the semantic-search, Chroma-retrieval, and RAG-evaluation electives.
- •Building a chat UI for a generative AI app with Gradio, without writing frontend code.
- •Fine-tuning an open-source LLM on your own data locally with the lamini framework, covered in one elective course.
Strengths
- ✓Courses are explicitly graded into required and elective, so a beginner has one clear starting sequence instead of 11 equally-weighted options.
- ✓Prompts were independently tested in Chinese rather than machine-translated, according to the README's own account of the process.
- ✓Every course ships as a runnable notebook, not just slides or a video transcript.
- ✓Covers the full pipeline end to end - prompting, app-building, RAG, and fine-tuning - instead of stopping at prompt engineering.
学习前提与局限性
- △The license is CC BY-NC-SA 4.0, so per the README, commercial use of this content isn't permitted without separate permission - fine for self-study, a problem if you wanted to fold it into a paid course.
- △It assumes you already have an LLM API key, preferably OpenAI's; the README says using a different provider means adapting the sample code yourself, and points to Datawhale's sibling project llm-universe for that case instead of covering it here.
- △Bilingual subtitles and a walkthrough video only exist for the Prompt Engineering course - the other ten courses are notebook-and-text only.
- △It's translation-and-reproduction work, not original curriculum design: the course structure and progression are Andrew Ng's, and llm-cookbook's own contribution is the Chinese-language layer, the reproduced code, and the required/elective grading.
- △GitHub shows no license field at the repo level; the CC BY-NC-SA terms only appear in the README text, so tooling that reads repo metadata directly may miss it.
同类学习资源对比
Frequently asked questions
LLM Cookbook is free to read online or download as a PDF, but the README licenses the content under CC BY-NC-SA 4.0 (Creative Commons Attribution-NonCommercial-ShareAlike). That means commercial use isn't permitted without separate permission from the Datawhale team; personal study and non-commercial sharing are fine.
LLM Cookbook expects basic Python ability and comfort running Jupyter notebooks, plus access to an LLM API key - the README recommends OpenAI's. If you're using a different provider, you'll need to adapt the sample code yourself, or check Datawhale's sibling project llm-universe, which the README points to for that case.
The required and elective courses cover three areas: prompting technique (two courses, including one on the ChatGPT API), app building (two LangChain courses plus one on Gradio), and model work (evaluation, debugging, finetuning, and four retrieval/RAG courses) — all reproduced from Andrew Ng's DeepLearning.AI catalog, translated into Chinese.
Only the Prompt Engineering course does. The README links a separately produced bilingual-subtitle video on Bilibili, a bilingual subtitle download from a companion GitHubDaily repo, and one extra video walkthrough. The other ten courses are text-and-notebook only, with no video version.
LangChain gets two required courses - building applications with the framework, then querying your own data with it - plus an elective on Functions, Tools, and Agents. RAG is split across four elective courses covering semantic search, Chroma-based retrieval, full RAG evaluation, and advanced retrieval techniques, building from basic to production-level pipelines.
Start with the four required courses in the README's listed order: prompt engineering, building a ChatGPT Q&A system, then the two LangChain courses, since each builds skills the next assumes. After that, the eight elective courses are meant to be picked based on interest rather than followed in sequence; the README doesn't prescribe an order for them.
Who should try it — and who should skip
Developers who already write Python, have (or can get) an LLM API key, and would rather work through a graded Chinese-language curriculum than piece together English blog posts and course transcripts should start with the four required courses. Skip it if you don't read Chinese - the original English DeepLearning.AI courses cover the same material - or if you're after LLM internals and pretraining, which this curriculum doesn't touch.
Related repositories
Curious whether llm-cookbook is right for you?
Let ChatGPT, Claude, or Perplexity look into it — click below and see what AI actually says about llm-cookbook.