Dive into LLMs

University-grade open curriculum for understanding (not just using) LLMs

Dive into LLMs is a structured LLM programming tutorial series from Shanghai Jiao Tong University covering fine-tuning, RLHF alignment, RAG pipelines, jailbreak attacks and defenses, watermarking techniques, GUI agents, and multimodal models. Each module includes slides, Jupyter notebooks with runnable code, and accompanying video lectures. The curriculum is designed for developers and researchers who want to go beyond prompt engineering into actually understanding how large language models work, how they're trained, and how to modify and deploy them. Topics span from transformer fundamentals through modern alignment techniques like DPO and GRPO. Recent additions cover GUI agents and multimodal architectures. The course has partnered with Huawei's Ascend community for localized deployment content. With 29k+ GitHub stars and trending hard today, this is one of the most-starred educational resources in the LLM ecosystem. Unlike blog posts and YouTube tutorials, the Jupyter notebooks mean you can run and modify every example yourself — making abstract concepts like RLHF tangible in a way that passive reading can't match.

Panel Reviews

The Builder

Developer Perspective

Ship

“Every dev who uses LLMs in production should understand fine-tuning and alignment at the level this curriculum teaches. The Jupyter notebooks are the key — being able to run RLHF examples on a small model changes your mental model for how alignment actually works.”

The Skeptic

Reality Check

Skip

“There are dozens of LLM curricula on GitHub — fast.ai, Andrej Karpathy's videos, the Stanford CS224N lectures. Unless you specifically need SJTU's framing or the Huawei Ascend content, it's hard to argue this is uniquely worth your time over the better-known alternatives.”

The Futurist

Big Picture

Ship

“The world needs millions more people who understand LLMs at the fine-tuning and alignment level — not just the API level. Open curricula like this are how that happens. The jailbreak and watermarking modules are especially forward-looking for an increasingly adversarial AI landscape.”

The Creator

Content & Design

Skip

“This is squarely for researchers and ML engineers, not creative practitioners. I appreciate the effort but nothing here helps me do my work better today — it's a long-form learning investment that most creators won't need to make.”