A Chinese-language Transformer boot camp built by students, for students

What it does

This repo is a four-chapter Chinese-language course that walks beginners from “what is attention?” to fine-tuning BERT on real tasks. It targets people who know Python and PyTorch but have never touched transformers. Each chapter ends with a quiz, which is a nice touch for a self-paced repo.

The interesting bit

The project is essentially a curated remix: it pulls from Hugging Face docs, visual explainers, and the authors’ own lecture notes, then stitches them into a single linear path. The authors are undergrads and grad students from Peking, Zhejiang, and McGill—so the tone is peer-to-peer rather than professorial. Chapter 3 actually builds BERT from scratch in PyTorch before letting you import BertModel in Chapter 4.

Key highlights

• Pure Chinese prose and comments, including the “illustrated” explainers for attention, Transformer, BERT, and GPT
• Hands-on task coverage: text classification, NER, extractive QA, multiple-choice QA, language modeling, translation, and summarization
• Self-contained quizzes at the end of each chapter to check understanding
• Explicitly cites its sources rather than pretending originality
• Maintained by Datawhale, a Chinese open-source learning collective

Caveats

• The repo is documentation-heavy; the “Shell” language tag is misleading—this is Markdown and notebooks, not a runnable framework
• No visible CI, issue triage, or recent commit history in the README; freshness is unclear
• You will need to bring your own GPU or Colab patience for the larger tasks

Verdict

Worth bookmarking if you are a Chinese-speaking beginner who finds the official Hugging Face docs too reference-dense. Skip it if you already know how to subclass PreTrainedModel or if you need a production-ready toolkit rather than a textbook.