What it does This is a collection of Jupyter notebooks, PDF slides, and README guides spun out of two SJTU courses (NIS8021 and NIS3353). It walks through practical LLM tasks: fine-tuning and deployment, prompt engineering, knowledge editing, math reasoning distillation, watermarking, jailbreak attacks, steganography, multimodal models, GUI agents, agent safety, and RLHF alignment. A newer “full workflow” track, developed with Huawei’s Ascend team, adds PPTs, lab manuals, and video tutorials aimed at domestic (China-based) hardware.

The interesting bit The breadth is unusual for a free course repo. Most tutorials pick one lane—this one treats model attack (jailbreaks) and model defense (watermarking, safety alignment) as equally worthy of a lab session. The “invisible ink” steganography chapter and the GUI-agent “order takeout for me” exercise suggest the authors enjoy the weird corners of the field.

Key highlights

• 11 themed chapters, each with slides, a tutorial README, and a runnable notebook
• Recently added topics: math reasoning, GUI agents, RLHF safety alignment, steganography
• A separate Huawei Ascend-branded track with video and tiered (beginner/intermediate/advanced) content
• Explicitly free and “public welfare” (公益性质); no paywall or institutional gate
• Version tagged v0.1.0, status marked “building”—actively accepting PRs and issues

Caveats

• All content is caveat emptor: the disclaimer says it comes from “personal experience, internet data, and daily research accumulation” with no guarantee of correctness
• The Ascend-specific track is hosted off-repo on Huawei’s community site; the GitHub side is mostly the original SJTU notebooks
• Chinese-language materials throughout; English speakers will need translation tools

Verdict Worth bookmarking if you want a structured, notebook-first survey of modern LLM engineering and security topics—especially if you read Chinese. Skip if you need production-grade reference implementations or English documentation.