nndl/nndl: Qiu Xipeng's Neural Networks and Deep Learning, Two Editions in One Repository
邱锡鹏《神经网络与深度学习》第二版与通识版：电子书、章节目录、学习资源与勘误。
At a glance
- What is it?
- The nndl/nndl repository is the distribution point for the second edition and the general-reader edition of Qiu Xipeng's Chinese deep learning textbook, plus errata tracking and links to PyTorch practice code. It is a book repository, not a library.
- Who is it for?
- Adopt nndl/nndl if you read Chinese and want a structured theory text with a PyTorch companion repository; the second edition assumes linear algebra, calculus and probability, so readers without that background should start with the general-reader edition and its stories and analogies. Skip it if you want runnable code in the same repository, because the LaTeX source is not public and the practice code lives in a separate project.
- Can I use it commercially?
- Not without permission. GitHub finds no licence file in the repository, and without a licence all rights are reserved by default: you may read the code but not reuse it. Check the README, or ask the authors, before using it.
- Is it still maintained?
- Yes. The repository last received commits 24 days ago.
- What is it written in?
- GitHub does not report a main language for this repository.
Answers come from the project's GitHub data, last synced on September 29, 2026, and from our analysis. They are not legal advice.
Editorial analysis
What nndl/nndl Actually Distributes
This is not a software project. The repository is the public face of two Chinese-language textbooks by Qiu Xipeng: the second edition of Neural Networks and Deep Learning (the so-called dandelion book) and a general-reader edition of the same title. What lives here is the README that routes readers to the right book, the chapter tables of contents, cover assets, a legacy folder for the first edition, and metadata directories named nndl-v2/ and nndl-ge/. The actual PDFs are attached to GitHub releases rather than committed to the tree, and the README points at https://nndl.ai/ as the series home. The audience is split deliberately: the second edition targets professional courses and the start of research work, while the general-reader edition targets non-specialists, cross-disciplinary beginners and university general-education courses. If you arrived expecting importable modules, you are in the wrong repository, and the README says so indirectly by never mentioning an install step for a library.
How the Two Editions Differ in Structure and Prerequisites
The second edition runs 16 chapters plus 5 mathematical appendices covering linear algebra, calculus, mathematical optimization, probability theory and information theory. Its three parts move from machine learning basics (linear models, feedforward networks, convolutional and recurrent networks) through network optimization, attention and Transformers, and graph neural networks, then into unsupervised learning, deep reinforcement learning, large language models and agents, probabilistic graphical models, deep belief networks and deep generative models. The general-reader edition also has 16 chapters, but the README states plainly that its four groups are browsing navigation, not formal divisions of the book. Its chapters trade derivations for stories and analogies, running from how machines learn on their own through attention, self-supervised learning, reinforcement learning, large language models, AI agents, diffusion models, multimodal AI, AI in science, embodied intelligence and AI risk. The prerequisite gap is the real dividing line: the README says the second edition needs linear algebra, calculus and probability, and that the appendices exist for review. The general-reader edition carries no such requirement.
Downloading the Second Edition PDF and Reading the First Chapter
There is nothing to compile. The README links each book's PDF directly to a release asset, and the most recent release is tagged book-pdf, dated 2026-07-23. To fetch the second edition from a shell, use the release URL the README gives.
curl -L -o nndl-v2.pdf \
https://github.com/nndl/nndl/releases/download/book-pdf/nndl-v2.pdfThe file that lands on disk is the full second-edition PDF. For the general-reader edition, the README gives a parallel URL ending in nndl-ge.pdf, so the same command with that filename produces the lighter book. If you prefer not to use the command line, the README's book table has a 下载 PDF link for each edition that resolves to the same release assets. One caution the README itself raises: both books are described as open pre-publication electronic drafts whose content changes with revisions, and it asks readers citing or giving feedback to note the version and download date. Save the download date alongside the file, because a citation to chapter 8 of the second edition is only meaningful if you record which revision you read.
Errata, Feedback and the Limits of This Repository
The README routes all corrections through GitHub Issues and asks contributors to search for an existing report first. A usable report includes the book title and version, the chapter and either the PDF page number or the printed page number, the PDF download or reading date, and the original text alongside the problem and a suggested fix. That template matters because the repository holds no source text to diff against: the maintenance notes state that the second edition's LaTeX source is not in this public repository, and that nndl-v2/ and nndl-ge/ contain only site metadata and description pages. The practical consequence is that you cannot send a pull request editing a paragraph. You can only file an issue and wait for a revision. The _meta.yml files inside each book directory feed the series site's book cards and are aggregated by a script in a separate repository, nndl.github.io, so even the metadata pipeline is not self-contained here. Treat this repository as a distribution and feedback channel, not a collaboration surface for the manuscript.
Practice Code Lives in nndl-practice, Not Here
The README's reading advice recommends pairing theory study with the second edition's PyTorch practice in the nndl/nndl-practice repository, which is where you verify models and algorithms by running them. The first edition's PaddlePaddle code sits in a separate repository as well. There is also a companion reader, Large Language Models and Agents, in nndl/llm-beginner, described as a topic reader with six standalone introductory exercises. So the study loop the project intends is split across three or four repositories: read a chapter from the PDF, then open a different repository for code, then file an issue here if the text is wrong. Anyone evaluating this project should weigh that fragmentation. It keeps the book repository small and the PDF the single artifact, but it means a newcomer following the README will switch contexts several times before running a single training step. The README does not document a rollback path for a bad revision, only that revisions happen.
Where This Textbook Fits Against Michael Nielsen's Book
The related searches around this project mix it with Michael Nielsen's Neural Networks and Deep Learning, and the confusion is understandable because both are free online books with deep learning in the title. The difference is language and scope. Nielsen's book is English, built around a small set of concrete examples and code you write yourself as you read, and it predates the Transformer era. Qiu Xipeng's second edition is Chinese, 16 chapters plus five math appendices, and its later chapters cover Transformers, graph neural networks, large language models and agents, diffusion models and deep reinforcement learning. The general-reader edition goes further into AI safety, embodied intelligence and scientific applications. If your goal is to implement backpropagation from scratch in Python while reading, Nielsen's approach fits better. If your goal is a systematic Chinese-language treatment that reaches current architectures and you already have the calculus and linear algebra, the second edition covers ground Nielsen's book does not attempt. They are complements, not substitutes.
Licence, Maintenance and Upgrade Cost
The repository metadata does not state a licence, and the README does not discuss one either. What the README does say is that both books are open pre-publication electronic drafts, that content updates with revisions, and that readers should note the version and download date when citing or giving feedback. That is a statement about reading and attribution, not a grant of reuse rights, so anyone planning to redistribute the PDFs, translate them or build course materials on top should confirm terms with the author rather than assume an open licence from the word 开放. On maintenance, the last push to the default branch was on 2026-09-06, and the most recent release, tagged book-pdf, dates from 2026-07-23. The repository is not archived. The upgrade cost is low in tooling terms, since there is no dependency to bump, but non-zero in citation terms: a chapter number can point at different text across revisions, so pin the release tag you downloaded.
Editorial conclusion
Adopt nndl/nndl if you read Chinese and want a structured theory text with a PyTorch companion repository; the second edition assumes linear algebra, calculus and probability, so readers without that background should start with the general-reader edition and its stories and analogies. Skip it if you want runnable code in the same repository, because the LaTeX source is not public and the practice code lives in a separate project. Before relying on any chapter, check the release date of the PDF you downloaded against the repository's issue list, since the README describes both books as pre-publication drafts that change with revisions.
Frequently asked questions
Is a neural network the same thing as deep learning?
The second edition treats neural networks as the model family and deep learning as the broader field built on them, moving from machine learning basics through feedforward, convolutional and recurrent networks before reaching Transformers and large language models. The general-reader edition explains the same progression through stories and analogies for readers without a mathematics background.
How do you explain a deep neural network to a beginner?
The README positions the general-reader edition as the entry point for exactly this audience, describing it as using stories, cases and everyday analogies to explain neural networks and deep learning. Its first four chapters cover how machines learn on their own and how a network is trained, before moving to attention and large language models.
What are the main types of neural networks covered in the nndl/nndl books?
The second edition's table of contents lists feedforward networks, convolutional networks, recurrent networks, graph neural networks, deep belief networks and deep generative models, with attention and Transformer as a dedicated chapter. The general-reader edition covers the same families at an intuitive level, adding diffusion models and multimodal AI.
Is ChatGPT a neural network?
The second edition includes a chapter on large language models and agents, and the general-reader edition devotes a chapter to the language model story, so the books treat systems like ChatGPT within the neural network and Transformer lineage they describe. Neither the README nor the tables of contents make a claim beyond that framing.
Official sources
Add this badge to your README
If you maintain this project, the badge below links readers to this analysis and shows its maintenance status from the daily GitHub snapshot. Paste the markdown into your README; add ?metric=license or ?metric=stars to the image URL for a different field.
[](https://hysenlabs.com/projects/nndl-nndl)