nndl/nndl-practice

What it solves

This project bridges the gap between theoretical understanding of deep learning and practical engineering. It helps learners move beyond abstract concepts by implementing core neural network components from scratch using PyTorch and applying them to real, runnable tasks.

How it works

The repository is organized into 10 chapters that mirror a textbook structure. Each chapter uses Jupyter Notebooks to guide you through implementing algorithms—starting from basic linear models and progressing to complex architectures like Transformers, Graph Neural Networks, and Large Language Models. As you progress, the project helps you build a reusable, lightweight training framework (nndl/) that evolves from simple training loops to professional-grade model management.

Who it’s for

It is designed for students and engineers who have a foundational knowledge of Python and want to learn deep learning through hands-on practice. It is intended to be used alongside a theoretical textbook.

Highlights

• Step-by-step implementation: Build key components like convolution, attention mechanisms, and optimization algorithms from the ground up.
• Progressive framework: Develop a modular, reusable code library (nndl/) as you advance through the chapters.
• Comprehensive coverage: Topics range from basic regression and neural networks to cutting-edge LLMs, agents, and graph neural networks.
• Verification tools: Includes sanity tests for each chapter to ensure your implementation matches expected model shapes and core logic.
• Flexible learning paths: Offers curated study tracks tailored to specific interests like computer vision, sequence modeling, or LLM development.