bbruceyuan/LLMs-Zero-to-Hero

What it solves

This project provides a comprehensive, hands-on learning path for understanding and building Large Language Models (LLMs) from the ground up. It bridges the gap between theoretical knowledge and practical implementation, helping learners transition from beginners to proficient practitioners.

How it works

The project uses a "learn-by-doing" approach, featuring step-by-step code implementations in Jupyter Notebooks that mirror the development of modern AI architectures. It covers the entire lifecycle of LLMs—including pre-training, supervised fine-tuning (SFT), DPO, and RLHF—while providing accompanying video tutorials and blog posts to explain the underlying concepts.

Who it’s for

It is designed for developers and AI enthusiasts who want to master LLM internals. It is particularly useful for those who prefer learning through code and practical experimentation rather than just reading academic papers.

Highlights

• Systematic curriculum covering everything from basic model architecture to advanced training techniques like MOE and DeepSeek-MLA.
• Fully functional, hands-on code examples that can be run on consumer-grade hardware (RTX 3090/4090).
• Integrated video tutorials and detailed blog explanations for every major topic.
• Practical guidance on model deployment, inference optimization, and quantization.
• A structured repository that serves as a foundation for an upcoming book on LLM development.