Dive into Deep Learning — Review & Reading Guide
An original review of the free, interactive deep-learning textbook by Zhang, Lipton, Li and Smola — who should read it, in what order, and what to skip.
This page replaces the original Engatica ebook listing with what it should always have been: a review, not a download. We do not host the book; it is freely available from its authors at d2l.ai.
What it is
Dive into Deep Learning (d2l.ai) is a full textbook that teaches deep learning as runnable code — every concept arrives with an implementation you can execute in PyTorch, JAX, TensorFlow or MXNet. It grew out of courses taught at Berkeley and elsewhere and is maintained by researchers including Zachary Lipton and Mu Li.
Who should read it
Engineers moving into ML who learn by building, and practitioners who want the maths and the code on the same page. Work through the preliminaries and linear models, then jump to the transformer chapters — in 2027 those are the load-bearing ones for anyone working with LLMs.
Our verdict
Still the best free route from zero to genuinely understanding the machinery your agents run on. Pair it with our LLM hub for the applied layer.
2027 update
Since this review first ran, the book has kept moving — which is the point of it. The live edition at d2l.ai is now version 1.0.3, still free and open-source, with every example runnable in PyTorch, JAX, TensorFlow and MXNet, and the authors report adoption at more than 500 universities in over 70 countries [1][2]. A second volume is in preparation, with specialist contributors leading individual chapters [2] — a sign the authors intend the transformer-era material to get the same treatment the fundamentals did. Two practical notes for 2027 readers. First, the Cambridge University Press print edition is a snapshot: the online text has continued to change since, so cite the web version when precision matters. Second, our original advice stands and has strengthened — work the preliminaries and linear models properly, then go straight to the attention and transformer chapters; in the agent era those are the load-bearing ones, because every framework you will use is a thin wrapper around them.
Frequently asked questions
Is Dive into Deep Learning free?
Yes, completely. The full book is free and open-source online at d2l.ai, and every example is executable code you can run in PyTorch, JAX, TensorFlow or MXNet. A Cambridge University Press print edition exists if you want paper, but the web version is the real book: it is updated continuously, while print is a snapshot of the text at one moment in time.
Which edition is current?
The live online edition, version 1.0.3 — there is no second edition to hunt for, because the authors version the web text instead. The project reports use at more than 500 universities in over 70 countries, and a second volume, with specialist contributors leading individual chapters, is in preparation. Until it lands, the practical advice is unchanged: cite the web version when precision matters, not the print snapshot.
Who is Dive into Deep Learning for?
Engineers moving into machine learning who learn by building, and practitioners who want the mathematics and the code on the same page. It assumes comfort with Python and basic linear algebra, then earns its keep: work the preliminaries and linear models properly, then go straight to the attention and transformer chapters. In the agent era those chapters are load-bearing — every framework you will use wraps them.
What should a 2027 reader pair it with?
Start one step earlier if the maths is rusty: our neural networks reading guide sequences the intuition videos and build-it-by-hand courses that make this book's opening chapters far less steep. Then, once the transformer chapters land, move to applied LLM material — agent frameworks and evals — because D2L deliberately teaches the machinery underneath them rather than the tools themselves.
Sources
- Dive into Deep Learning, version 1.0.3 (live edition) — Zhang, Lipton, Li & Smola (2026)
- Dive into Deep Learning (project site: 500+ universities; second volume contributors; four frameworks) — Dive into Deep Learning project (2026)
- Dive into Deep Learning Adds Attention Mechanism Chapter — Amazon Science (2024)