A Chinese deep-learning textbook, now with PyTorch and LLM chapters

What it does

This repo holds runnable Jupyter notebooks that implement every chapter of 邱锡鹏’s Neural Networks and Deep Learning: Cases and Practice — a standard Chinese deep-learning textbook — in modern PyTorch. Ten chapters are marked “初版完成” (first draft complete), from linear models up through graph neural networks and, notably, a new chapter on large language models and agents. Each chapter ships with a README summarizing implementation points and pytest sanity checks.

The interesting bit

The repo is essentially a living textbook appendix. It replaced the older nndl/exercise repository (now stashed in legacy/) and tracks an upcoming 2026 print edition rather than lagging behind it. The maintainers also publish a parallel PaddlePaddle version, so the PyTorch release here is the more broadly useful fork for most developers.

Key highlights

• 10 chapters covering fundamentals → CNNs → RNNs → attention → GNNs → LLMs/agents
• Every chapter: notebooks + explanatory README + automated sanity tests
• Built to accompany a printed textbook, not a loose blog-tutorial series
• Stars and history carried over from the 2017–2024 predecessor repo
• Part of a larger ecosystem: theory book, general-audience edition, and a separate LLM beginner track

Caveats

• All materials are in Chinese; code comments and documentation assume Chinese literacy
• “初版完成” suggests first-draft completeness — polish and errata fixes may still be ongoing
• The 2026 print edition is not yet published, so chapter contents could still shift

Verdict

Worth bookmarking if you read Chinese and want textbook-grade, tested implementations that progress from logistic regression to LLM agents. Skip it if you need English explanations or are hunting for novel research code rather than pedagogical fundamentals.