TinyTorch/docs/students/project-guide.md

🔥 TinyTorch Project Guide

Building Machine Learning Systems from Scratch

This guide helps you navigate through the complete TinyTorch course. Each module builds progressively toward a complete ML system using a notebook-first development approach with nbdev.

🎯 Module Progress Tracker

Track your progress through the course:

• Module 0: Setup - Environment & CLI setup
• Module 1: Tensor - Core tensor operations
• Module 2: MLP - Multi-layer perceptron
• Module 3: CNN - Convolutional networks
• Module 4: Autograd - Automatic differentiation
• Module 5: Data - Data loading pipeline
• Module 6: Training - Training loop & optimization
• Module 7: Config - Configuration system
• Module 8: Profiling - Performance profiling
• Module 9: Compression - Model compression
• Module 10: Kernels - Custom compute kernels
• Module 11: Benchmarking - Performance benchmarking
• Module 12: MLOps - Production monitoring

🚀 Getting Started

First Time Setup

1. Clone the repository
2. Go to: modules/setup/README.md
3. Follow all setup instructions
4. Verify with: python modules/setup/check_setup.py

Daily Workflow

cd TinyTorch
source .venv/bin/activate  # Always activate first!
python bin/tito.py info            # Check system status

📋 Module Development Workflow

Each module follows this pattern:

1. Read overview: modules/[name]/README.md
2. Work in notebook: modules/[name]/[name]_dev.ipynb
3. Export code: python bin/tito.py sync
4. Run tests: python bin/tito.py test --module [name]
5. Move to next module when tests pass

📚 Module Details

🔧 Module 0: Setup

Goal: Get your development environment ready Time: 30 minutes Location: modules/setup/

Key Tasks:

• Create virtual environment
• Install dependencies
• Implement hello_tinytorch() function
• Pass all setup tests
• Learn the tito CLI

Verification:

python modules/setup/check_setup.py  # Should show all ✅
python bin/tito.py test --module setup

---

🔢 Module 1: Tensor

Goal: Build the core tensor system Prerequisites: Module 0 complete Location: modules/tensor/

Key Tasks:

• Implement Tensor class in notebook
• Basic operations (add, mul, reshape)
• Memory management
• Shape validation
• Broadcasting support

Verification:

python bin/tito.py test --module tensor

---

🧠 Module 2: MLP

Goal: Build multi-layer perceptron Prerequisites: Module 1 complete Location: modules/mlp/

Key Tasks:

• Implement Linear layer
• Activation functions (ReLU, Sigmoid)
• Forward pass
• Basic backward pass
• Train on MNIST

Target: >95% accuracy on MNIST

---

🖼️ Module 3: CNN

Goal: Build convolutional neural networks Prerequisites: Module 2 complete Location: modules/cnn/

Key Tasks:

• Implement Conv2d layer
• MaxPool2d layer
• Padding and stride support
• Train CNN on CIFAR-10

Target: >80% accuracy on CIFAR-10

---

⚡ Module 4: Autograd

Goal: Automatic differentiation engine Prerequisites: Module 3 complete Location: modules/autograd/

Key Tasks:

• Computational graph construction
• Backward pass automation
• Gradient checking
• Memory efficient gradients

Verification: All gradient checks pass

---

📊 Module 5: Data

Goal: Efficient data loading Prerequisites: Module 4 complete Location: modules/data/

Key Tasks:

• Custom DataLoader implementation
• Batch processing
• Data transformations
• Multi-threaded loading

---

🎯 Module 6: Training

Goal: Complete training system Prerequisites: Module 5 complete Location: modules/training/

Key Tasks:

• Training loop implementation
• SGD optimizer
• Adam optimizer
• Learning rate scheduling
• Metric tracking

---

⚙️ Module 7: Config

Goal: Configuration management Prerequisites: Module 6 complete Location: modules/config/

Key Tasks:

• YAML configuration system
• Experiment logging
• Reproducible training
• Hyperparameter management

---

📊 Module 8: Profiling

Goal: Performance measurement Prerequisites: Module 7 complete Location: modules/profiling/

Key Tasks:

• Memory profiler
• Compute profiler
• Bottleneck identification
• Performance visualizations

---

🗜️ Module 9: Compression

Goal: Model compression techniques Prerequisites: Module 8 complete Location: modules/compression/

Key Tasks:

• Pruning implementation
• Quantization
• Knowledge distillation
• Compression benchmarks

---

🔥 Module 10: Kernels

Goal: Custom compute kernels Prerequisites: Module 9 complete Location: modules/kernels/

Key Tasks:

• Optimized matrix multiplication
• Vectorized operations
• CPU optimization

---

📈 Module 11: Benchmarking

Goal: Performance benchmarking system Prerequisites: Module 10 complete Location: modules/benchmarking/

Key Tasks:

• Benchmark suite implementation
• Performance regression testing
• Comparative analysis
• Automated reporting

---

🚀 Module 12: MLOps

Goal: Production monitoring and deployment Prerequisites: Module 11 complete Location: modules/mlops/

Key Tasks:

• Model monitoring
• Production deployment
• A/B testing framework
• Performance dashboards

🧪 Testing Strategy

Module-Level Testing

# Test specific module
python bin/tito.py test --module tensor

# Test all modules
python bin/tito.py test

# Check overall status
python bin/tito.py info

Integration Testing

Each module integrates into the main tinytorch package through nbdev:

• Notebooks automatically export to tinytorch/core/
• Integration tests verify cross-module compatibility
• Final package assembly validates the complete system

🏆 Completion Criteria

A module is complete when:

• All notebook cells run without errors
• tito sync exports code successfully
• tito test --module [name] passes all tests
• Integration with previous modules works
• Ready to proceed to next module

🎓 Learning Philosophy

This course teaches systems thinking by building one cohesive ML system rather than isolated components. Each module contributes essential functionality that later modules depend upon, creating a complete, production-ready machine learning framework.

💡 Need Help?

• Quick Start: See modules/setup/QUICKSTART.md
• Development Workflow: Each module's README.md
• CLI Reference: python bin/tito.py --help
• Integration Issues: Check python bin/tito.py info