Overview
MicroGPT Course is an educational deep-dive that deconstructs Andrej Karpathy's minimal GPT implementation line-by-line, building understanding from fundamental math to a working language model.
Key Features
- 33 Comprehensive Lessons: Organized into 6 modules covering data, autograd, architecture, training, and inference
- No Prerequisites: Assumes only 10th-grade math and basic Python knowledge
- First-Principles Approach: Every concept explained from the ground up with no hand-waving
- Interactive Learning: Mermaid diagrams, KaTeX equations, and styled admonitions
Course Structure
- Data & Tokenization: Character-level encoding and vocabulary
- Calculus & Autograd: Custom automatic differentiation engine
- Architecture: Transformer with attention, MLP, and residual connections
- Training: Adam optimizer and backpropagation
- Inference: Temperature-controlled text generation
Technical Details
Built with MkDocs Material for an elegant reading experience. The course deconstructs a complete GPT model written in pure Python with zero external dependencies—proving that modern AI isn't magic, just math and careful engineering.