Shanghai Jiao Tong's free LLM course goes far beyond fine-tuning

What it does

This repo packages graduate coursework from Shanghai Jiao Tong University into a free, public set of hands-on LLM tutorials. Each of the eleven chapters ships with PDF slides, a written guide, and a runnable Jupyter notebook covering a distinct practical topic—ranging from vanilla fine-tuning and prompting to model watermarking, jailbreak attacks, and RLHF alignment. The goal is to get students and researchers writing code instead of just reading papers.

The interesting bit

While most tutorials hug the polite side of LLMs, this one spends equal time on adversarial corners: one chapter teaches you to embed invisible text watermarks, another shows how to jailbreak a model, and a third walks through steganography (“invisible ink”). It also recently launched a separate Huawei Ascend hardware track, porting the same hands-on approach to domestic Chinese AI silicon. The maintainers keep the tone direct—one RLHF tutorial warns it is “very dangerous” and asks you to check whether your model is sneering.

Key highlights

• Eleven chapters, each with slides, a README guide, and a .ipynb script
• Unusual breadth for an educational repo: knowledge editing, LLM steganography, GUI agents, and math-reasoning distillation (a “mini R1”)
• Backed by an active research team from SJTU and NUS, with a new Huawei Ascend community edition offering video lectures and lab manuals
• Explicitly free and public-welfare oriented, though it disclaims that tips come from personal experience and may not be 100% correct
• Version v0.1.0 and marked “building,” so the scope is ambitious and still expanding

Caveats

• The README flags the project as “building” and version v0.1.0, so expect rough edges and incomplete coverage
• Content is drawn from contributors’ personal experience and internet data rather than peer-reviewed guarantees, as the disclaimer notes
• Newer chapters and the Huawei Ascend track are recent additions, so maturity likely varies across topics

Verdict

Best for graduate students, researchers, or impatient developers who want a structured, code-first tour of LLM internals—including the shadowy corners. Skip it if you need a polished, drop-in framework or an English-first resource; the material is primarily in Chinese and aims at learning by doing, not production deployment.