# dive-into-llms is eleven course chapters and no licence file

> A free hands-on LLM course from Shanghai Jiao Tong University, expanded from two lecture courses, with each chapter shipping a slide deck, a tutorial and a runnable notebook. Four of the eleven chapters are attack-side material, the repository declares no licence, and the last push was on 2025-10-10.

**Lordog/dive-into-llms** — GitHub describes it as 《动手学大模型Dive into LLMs》系列编程实践教程. The repository metadata lists Jupyter Notebook as its primary language. This article stays within the project description and details documented in the GitHub repository README.

- Repository: https://github.com/Lordog/dive-into-llms
- Stars: 55,477 · Forks: 6,616
- Language: Jupyter Notebook
- License: not declared
- Published: 2026-08-13 · Updated: 2026-08-18 · Language: en
- Canonical page: https://hysenlabs.com/projects/lordog-dive-into-llms

## Eleven chapters, each a PDF, a tutorial and a notebook

The unit of this project is a chapter, and each chapter is built the same way three times over. Every entry in the table of contents links to slides, a PDF, a tutorial README and a script, and the script is always a Jupyter notebook. The order runs from fine-tuning and deployment through prompting and chain of thought, knowledge editing, mathematical reasoning, watermarking, jailbreak attacks, steganography, multimodal models, GUI agents, agent security and PPO-based RLHF alignment, which is chapter eleven. The last few are where the interesting material sits. It began as an expansion of two Shanghai Jiao Tong University courses, NIS8021 on frontier natural language processing techniques and NIS3353 on AI security techniques, and it is stated to be free and non-commercial in intent, aimed at helping students start on large model work for coursework or research.

## The repository is a documents folder and a README

Look at what is actually in the tree, because it changes what kind of artefact this is. The top level holds four entries: .gitignore, README.md, documents/ and pics/. There is no package directory, no setup script, no requirements file and no test suite. Everything is a document, and the chapter files live under documents/chapter1 through documents/chapter11 with the notebook and its PDF inside. That means there is no supported way to install any of it and nothing to pin a version of, so a reader who wants to reproduce a chapter clones the repository and runs the notebook in whatever environment they already have. The absence of a pinned environment is the practical gap, because a notebook that calls a hosted model API is exactly the kind of thing that breaks when an API version changes underneath it.

## Four chapters teach the attack side of model security

The security material is the part a reader should look at twice before assigning it. Chapter six is explicitly about jailbreak attacks and is framed with the argument that you cannot get better safety until you understand how the attack works. Chapter seven covers steganography, embedding information in model output that a human cannot see, and chapter five covers watermarking from the other direction, marking generated text so it can be identified. Chapter ten asks whether a large model agent realises risk in open agent settings, and chapter eleven is a PPO-based RLHF safety alignment experiment. The context is a university AI security course, which is a legitimate frame, but the artefacts are runnable notebooks rather than prose, so anyone using this as a syllabus is distributing working attack code and should decide deliberately what that means for the audience.

## The Ascend course is a link out, not a directory in

There is a second, larger course announced in the project, and it is easy to mistake it for part of this repository. The full-process LLM development tutorial was produced with Huawei's Ascend team and covers Ascend hardware and software, offered as slides, lab manuals and video across beginner, intermediate and advanced levels. The repository does not contain it. What it contains is a pointer to the Ascend community learning area where the series is hosted, and a note that it is aimed at researchers and developers working on the models Ascend already supports. So a reader looking for the runnable material for that second course has to leave this repository entirely, and a reader assessing scope should treat the eleven chapters here as the whole of what is on offer locally.

## No licence file, which is the real blocker for reuse

The repository facts record the licence as unknown, and that matches what the tree shows, since there is no LICENSE file alongside the README. That single absence decides what a reader can do with eleven chapters of course material. Absent a grant, there is no stated permission to copy the notebooks into a company training programme, to adapt a chapter for an internal course, or to redistribute the slides, and the default position under copyright is that none of those are available. This is not unusual for a university handout and it is not an accusation, but it means the material is effectively read-only. Anyone planning to build on it should resolve the licensing question with the contributors through the project's issue tracker before treating a chapter as a starting point for their own material.

## One release tag, and a last push on 2025-10-10

The project has a single release, tagged v1 and dated 2025-06-12, and the most recent push to the repository was on 2025-10-10, close to a year ago. That is too long ago to describe the project as still being worked on, and nothing in the tree suggests otherwise. There is a dated update note from 2025-06-06 recording what changed in that release, namely the Ascend course going live and four new topics added to the original series, which are mathematical reasoning, GUI agents, model alignment and steganography. Reading the changelog rather than the branch history is the practical instruction here, because there is only the one entry and no per-chapter record of what was revised afterwards.

## The authors disclaim their own correctness

The project states its own limits, and they are worth quoting rather than paraphrasing. Everything in the tutorial comes from the contributors' personal experience, public internet data and their day-to-day research, and the text says the techniques are offered for reference and are not guaranteed to be one hundred percent correct. It asks for issues and pull requests where something is wrong, and it describes itself as a project still in progress where omissions are to be expected. That framing is honest and it also sets the reader's expectation correctly. This is lecture material that happens to be executable, not a maintained library with a compatibility promise, and treating it as the second would be the mistake.

## Conclusion

Adopt a chapter from here if you are teaching or self-studying and want a worked notebook per topic, particularly for the alignment and agent-security material that is hard to find organised anywhere else. Do not adopt it as a codebase, because there is nothing to install, nothing to depend on and no licence granting you reuse. Verify first whether the notebook in the chapter you care about still runs against current model APIs, since the repository itself disclaims correctness and its most recent push was on 2025-10-10, and read the notebook's own scope before assigning the attack-side chapters to students.

## FAQ

### What does the dive-into-llms tutorial cover?

Eleven chapters spanning fine-tuning and deployment, prompting and chain of thought, knowledge editing, mathematical reasoning, watermarking, jailbreak attacks, steganography, multimodal models, GUI agents, agent security and PPO-based RLHF alignment. It grew out of two Shanghai Jiao Tong University courses and is free.

### How is each chapter in dive-into-llms organised?

Each chapter links to three things, a slide deck as a PDF, a tutorial README, and a script which is a Jupyter notebook. The files sit under a per-chapter directory in the documents folder.

### Is the full-process LLM development course part of this repository?

No. The Ascend-based full-process course was built with Huawei and is hosted on the Ascend community learning area, and the repository only links out to it. The eleven chapters are the whole of what is present locally.

### Can I reuse the dive-into-llms notebooks commercially?

Nothing in the repository grants that. The licence is recorded as unknown and there is no licence file, so the project states no permission to copy, adapt or redistribute the material, and the authors describe it as free and non-commercial in intent.

## Sources

- [Official README](https://github.com/Lordog/dive-into-llms#readme)
- [Project repository](https://github.com/Lordog/dive-into-llms)
- [Release notes](https://github.com/Lordog/dive-into-llms/releases)

---

Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/lordog-dive-into-llms