TinyTorch/site/STUDENT_QUICKSTART.md

🎓 TinyTorch Student Quickstart Guide

Welcome to TinyTorch! You're about to build an ML framework from scratch and understand ML systems engineering.

🚀 Getting Started (2 minutes)

1️⃣ Setup Your Environment

# Clone the repository
git clone https://github.com/MLSysBook/TinyTorch.git
cd TinyTorch

# One-command setup (handles everything!)
./setup-environment.sh

# Activate environment
source activate.sh

# Verify setup
tito system doctor

✅ You should see all green checkmarks!

What the setup script does:

• Creates virtual environment (optimized for your system)
• Installs all dependencies (NumPy, Jupyter, Rich, etc.)
• Configures TinyTorch for development
• Handles Apple Silicon architecture automatically

2️⃣ Start Your First Module

# View the first module
tito module view 01_tensor

# Or open the notebook directly
jupyter notebook modules/01_tensor/tensor_dev.py

📚 Learning Path

Module Progression

Each module builds on the previous one:

Module	What You'll Build	Capability Unlocked
01 Tensor	Core data structure	Manipulate ML building blocks
02 Tensor	Core data structure	Manipulate ML building blocks
03 Activations	Non-linearity functions	Add intelligence to networks
04 Layers	Neural network layers	Build network components
05 Dense	Complete networks	Create multi-layer networks
06 Spatial	Convolution operations	Process images
07 Attention	Attention mechanisms	Understand sequences
08 Dataloader	Data pipelines	Efficient data loading
09 Autograd	Automatic differentiation	Compute gradients
10 Optimizers	Training algorithms	Optimize networks
11 Training	Complete training loops	End-to-end learning
12 Compression	Model optimization	Deploy efficiently
13 Kernels	Custom operations	Hardware acceleration
14 Benchmarking	Performance analysis	Find bottlenecks
15 MLOps	Production systems	Deploy and monitor
16 TinyGPT	Language models	Build transformers

📝 Interactive Learning

Answer ML Systems Questions

Each module has 3 interactive questions where you write 150-300 word responses:

# %% nbgrader={"grade": true, "grade_id": "ml_systems_q1", ...}
"""
YOUR RESPONSE HERE

[Write your analysis about memory usage, scaling, or system design]
"""

Tips for Good Responses:

• Reference the actual code you implemented
• Discuss memory/performance implications
• Compare to production systems (PyTorch, TensorFlow)
• Think about scaling to larger models

🎯 Track Your Progress

Check Your Capabilities

# See overall progress
tito checkpoint status

# Visual timeline
tito checkpoint timeline

# Test specific capability
tito checkpoint test 01

Complete Modules

# When you finish a module, validate it
tito module complete 02_tensor

# This will:
# 1. Export your code to the package
# 2. Run integration tests
# 3. Test your capability checkpoint

💡 Learning Tips

1. Build First, Understand Through Building

Don't just read - type the code, run it, break it, fix it!

2. Test Immediately

After each implementation, run the test right away:

# Implementation
def my_function():
    return result

# Test immediately
assert my_function() == expected
print("✅ Test passed!")

3. Think About Systems

For every operation, ask:

• How much memory does this use?
• What's the time complexity?
• How would this scale to 1B parameters?
• What would break in production?

4. Use the Profiler

Many modules include profiling code:

with MemoryProfiler() as prof:
    result = operation()
print(prof.report())

🆘 Getting Help

Module Issues

# Check module status
tito module status

# Run tests for debugging
tito module test 02_tensor

# View detailed errors
tito module test 02_tensor --verbose

Environment Issues

# Full system check
tito system doctor

# Reset if needed
tito system reset

Community

• GitHub Issues: Report problems
• Discussions: Ask questions and share insights

🏆 Challenge Yourself

After Each Module

1. Run all tests successfully
2. Answer all ML Systems questions thoughtfully
3. Pass the capability checkpoint
4. Try modifying the code and see what breaks

Final Goal

Complete all 16 modules and build TinyGPT - a working language model using the framework you built!

📊 Submission (For Courses)

If you're taking this as a course:

Submit Your Work

# Your instructor will provide submission instructions
# Typically involves pushing to a specific repository
git add .
git commit -m "Complete module 02_tensor"
git push origin main

Grading

Your work is graded on:

1. Code Implementation (auto-graded)
2. Test Passing (auto-graded)
3. ML Systems Questions (manually graded)
4. Checkpoint Achievements (auto-validated)

---

Ready to build ML systems from scratch? Start with Module 01! 🚀

tito module view 01_setup