TinyTorch

Build ML Systems From First Principles

📢 December 2024 Release - TinyTorch is ready for community review! All 20 modules (Tensor → Transformers → Optimization → Capstone) are implemented with complete solutions. Seeking feedback on pedagogy, implementation quality, and learning progression. Student version tooling exists but is untested. This release focuses on validating the educational content before classroom deployment.

🎯 For Reviewers: Read the 📚 Jupyter Book to evaluate pedagogy. Clone the repo to run implementations. See INSTRUCTOR.md for classroom deployment and grading strategies.

📖 Table of Contents

• Why TinyTorch?
• What You'll Build - Including the CIFAR-10 North Star Goal
• Quick Start - Get running in 5 minutes
• Learning Journey - 20 progressive modules
• Learning Progression & Checkpoints - 21 capability checkpoints
• Key Features - Essential-only design
• Milestone Examples - Real achievements
• Documentation & Resources - For students, instructors, developers
• Ready to Start Building? - Your path forward

Why TinyTorch?

"Most ML education teaches you to use frameworks. TinyTorch teaches you to build them."

In an era where AI is reshaping every industry, the difference between ML users and ML engineers determines who drives innovation versus who merely consumes it. TinyTorch bridges this critical gap by teaching you to build every component of modern AI systems from scratch—from tensors to transformers.

A Harvard University course that transforms you from framework user to systems engineer, giving you the deep understanding needed to optimize, debug, and innovate at the foundation of AI.

What You'll Build

A complete ML framework capable of:

🎯 North Star Achievement: Train CNNs on CIFAR-10 to 75%+ accuracy

• Real computer vision with 50,000 training images
• Built entirely from scratch using only NumPy
• Competitive performance with modern frameworks

Additional Capabilities:

• Building GPT-style language models with attention mechanisms
• Modern optimizers (Adam, SGD) with learning rate scheduling
• Performance profiling, optimization, and competitive benchmarking
• Complete ML systems pipeline from tensors to deployment

No dependencies on PyTorch or TensorFlow - everything is YOUR code!

Repository Structure

TinyTorch/
├── src/               # 💻 Python source files (developers/contributors edit here)
│   ├── 01_tensor/        # Module 01: Tensor operations from scratch
│   │   ├── 01_tensor.py  # Python source (version controlled)
│   │   └── ABOUT.md      # Conceptual overview & learning objectives
│
├── modules/           # 📓 Generated notebooks (learners work here)
│   ├── 01_tensor/        # Auto-generated from src/
│   │   └── 01_tensor.ipynb  # Jupyter notebook for learning
│   │   ├── README.md     # Practical implementation guide
│   │   └── tensor.py     # Your implementation
│   ├── 02_activations/   # Module 02: ReLU, Softmax activations
│   ├── 03_layers/        # Module 03: Linear layers, Module system
│   ├── 04_losses/        # Module 04: MSE, CrossEntropy losses
│   ├── 05_autograd/      # Module 05: Automatic differentiation
│   ├── 06_optimizers/    # Module 06: SGD, Adam optimizers
│   ├── 07_training/      # Module 07: Complete training loops
│   ├── 08_dataloader/    # Module 08: Efficient data pipelines
│   ├── 09_spatial/       # Module 09: Conv2d, MaxPool2d, CNNs
│   ├── 10_tokenization/  # Module 10: Text processing
│   ├── 11_embeddings/    # Module 11: Token & positional embeddings
│   ├── 12_attention/     # Module 12: Multi-head attention
│   ├── 13_transformers/  # Module 13: Complete transformer blocks
│   ├── 14_profiling/     # Module 14: Performance analysis
│   ├── 15_quantization/  # Module 15: Model compression (precision reduction)
│   ├── 16_compression/   # Module 16: Pruning & distillation
│   ├── 17_memoization/   # Module 17: KV-cache/memoization
│   ├── 18_acceleration/  # Module 18: Hardware optimization
│   ├── 19_benchmarking/  # Module 19: Performance measurement
│   └── 20_capstone/      # Module 20: Complete ML systems
│
├── site/              # 🌐 Course website (Jupyter Book)
│   ├── intro.md          # Landing page
│   ├── _toc.yml          # Site navigation (links to modules)
│   └── chapters/         # Site-specific content
│
├── docs/              # 📚 PDF book generation
│   ├── _config_pdf.yml   # PDF-specific configuration
│   ├── _toc_pdf.yml      # Linear chapter ordering
│   ├── cover.md          # Book cover
│   └── preface.md        # Book preface
│
├── milestones/        # 🏆 Historical ML evolution - prove what you built!
│   ├── 01_1957_perceptron/   # Rosenblatt's first trainable network
│   ├── 02_1969_xor_crisis/   # Minsky's challenge & multi-layer solution
│   ├── 03_1986_mlp_revival/  # Backpropagation & MNIST digits
│   ├── 04_1998_cnn_revolution/ # LeCun's CNNs & CIFAR-10
│   ├── 05_2017_transformer_era/ # Attention mechanisms & language
│   └── 06_2024_systems_age/  # Modern optimization & profiling
│
├── tinytorch/         # 📦 Generated package (auto-built from your work)
│   ├── core/          # Your tensor, autograd implementations
│   ├── nn/            # Your neural network components
│   └── optim/         # Your optimizers
│
├── tests/             # 🧪 Comprehensive validation system
│   ├── 01_tensor/     # Per-module integration tests
│   ├── 02_activations/
│   └── ...            # Tests mirror module structure
│
└── tito/              # 🛠️ CLI tool for workflow automation
    ├── commands/      # Student/instructor workflow commands
    └── core/          # Core utilities

🚨 CRITICAL: Understand the Three Layers

• 📝 Source: Edit src/XX_name/XX_name.py (for contributors) OR work in generated modules/XX_name/XX_name.ipynb (for learners)
• 📓 Notebooks: Generated from source with tito export → creates modules/*.ipynb for learning
• 📦 Package: Import from tinytorch.core.component → auto-generated from notebooks
• ❌ Never edit: Files in tinytorch/ directly (regenerated on every export)
• 🔄 Workflow: src/*.py → modules/*.ipynb → tinytorch/*.py

Why this structure?

• Developers: Edit Python source (src/) for version control
• Learners: Work in notebooks (modules/) for interactive learning
• Both: Import completed components from tinytorch/ package

Quick Start

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

# One-command setup (handles Apple Silicon, dependencies, everything)
./setup-environment.sh

# Activate environment
source activate.sh

# Verify setup
tito system health

# Start building
tito module start 01_tensor

That's it! The setup script handles:

• ✅ Virtual environment creation (arm64 on Apple Silicon)
• ✅ All required dependencies (NumPy, Rich, PyYAML, pytest, jupytext)
• ✅ TinyTorch package installation in development mode
• ✅ Architecture detection and optimization

Optional Dependencies (for visualization and advanced features):

• matplotlib - For plotting in Modules 17, 19, 20 (optional)
• jupyter - For interactive development (optional)

Note: Memory profiling uses Python's built-in tracemalloc module (standard library). System information uses os.cpu_count() and platform module (standard library). No external system monitoring dependencies required!

Learning Journey

20 Progressive Modules

Part I: Neural Network Foundations (Modules 1-7)

Build and train neural networks from scratch

Module	Topic	What You Build	ML Systems Learning
01	Tensor	N-dimensional arrays + operations	Memory layout, cache efficiency, broadcasting semantics
02	Activations	ReLU + Softmax (essential functions)	Numerical stability, gradient flow, function properties
03	Layers	Linear layers + Module abstraction	Parameter management, weight initialization, forward/backward
04	Losses	MSE + CrossEntropy (essential losses)	Numerical precision, loss landscapes, training objectives
05	Autograd	Automatic differentiation engine	Computational graphs, memory management, gradient flow
06	Optimizers	SGD + Adam (essential optimizers)	Memory efficiency (Adam uses 3x memory), convergence
07	Training	Complete training loops + evaluation	Training dynamics, checkpoints, monitoring systems

Milestone Achievement: Train XOR solver and MNIST classifier after Module 7

---

Part II: Computer Vision (Modules 8-9)

Build CNNs that classify real images

Module	Topic	What You Build	ML Systems Learning
08	DataLoader	Efficient data pipelines + CIFAR-10	Batch processing, memory-mapped I/O, data pipeline bottlenecks
09	Spatial	Conv2d + MaxPool2d + CNN operations	Parameter scaling, spatial locality, convolution efficiency

Milestone Achievement: CIFAR-10 CNN with 75%+ accuracy

---

Part III: Language Models (Modules 10-14)

Build transformers that generate text

Module	Topic	What You Build	ML Systems Learning
10	Tokenization	Text processing + vocabulary	Vocabulary scaling, tokenization bottlenecks, sequence processing
11	Embeddings	Token embeddings + positional encoding	Embedding tables (vocab × dim parameters), lookup performance
12	Attention	Multi-head attention mechanisms	O(N²) scaling, memory bottlenecks, attention optimization
13	Transformers	Complete transformer blocks	Layer scaling, memory requirements, architectural trade-offs
14	Profiling	Performance analysis + bottleneck detection	Memory profiling, FLOP counting, Amdahl's Law, performance measurement

Milestone Achievement: TinyGPT language generation with optimized inference

---

Part IV: System Optimization (Modules 15-20)

Profile, optimize, and benchmark ML systems

Module	Topic	What You Build	ML Systems Learning
14	Profiling	Performance analysis + bottleneck detection	Memory profiling, FLOP counting, Amdahl's Law, performance measurement
15	Quantization	Model compression + precision reduction	Precision trade-offs (FP32→INT8), memory reduction, accuracy preservation
16	Compression	Pruning + knowledge distillation	Sparsity patterns, parameter reduction, compression ratios
17	Memoization	Computational reuse via KV-caching	Memory vs compute trade-offs, cache management, generation efficiency
18	Acceleration	Hardware optimization + cache-friendly algorithms	Cache hierarchies, memory access patterns, vectorization vs loops
19	Benchmarking	Performance measurement + TinyMLPerf competition	Competitive optimization, relative performance metrics, innovation scoring
20	Capstone	Complete end-to-end ML systems project	Integration, production deployment, real-world ML engineering

Milestone Achievement: TinyMLPerf optimization competition & portfolio capstone project

---

Learning Philosophy

Most courses teach you to USE frameworks. TinyTorch teaches you to UNDERSTAND them.

# Traditional Course:
import torch
model.fit(X, y)  # Magic happens

# TinyTorch:
# You implement every component
# You measure memory usage
# You optimize performance
# You understand the systems

Why Build Your Own Framework?

• Deep Understanding - Know exactly what loss.backward() does
• Systems Thinking - Understand memory, compute, and scaling
• Debugging Skills - Fix problems at any level of the stack
• Production Ready - Learn patterns used in real ML systems

Learning Progression & Checkpoints

Capability-Based Learning System

Track your progress through capability-based checkpoints that validate your ML systems knowledge:

# Check your current progress
tito checkpoint status

# See your capability development timeline
tito checkpoint timeline

Module Progression:

• 01-02: Foundation (Tensor, Activations)
• 03-07: Core Networks (Layers, Losses, Autograd, Optimizers, Training)
• 08-09: Computer Vision (DataLoader, Spatial ops - unlocks CIFAR-10 @ 75%+)
• 10-13: Language Models (Tokenization, Embeddings, Attention, Transformers)
• 14-19: System Optimization (Profiling, Memoization, Quantization, Compression, Acceleration, Benchmarking)
• 20: Capstone (Complete end-to-end ML systems)

Each module asks: "Can I build this capability from scratch?" with hands-on validation.

Module Completion Workflow

# Complete a module (automatic export + testing)
tito module complete 01_tensor

# This automatically:
# 1. Exports your implementation to the tinytorch package
# 2. Runs the corresponding capability checkpoint test
# 3. Shows your achievement and suggests next steps

Key Features

Essential-Only Design

• Focus on What Matters: ReLU + Softmax (not 20 activation functions)
• Production Relevance: Adam + SGD (the optimizers you actually use)
• Core ML Systems: Memory profiling, performance analysis, scaling insights
• Real Applications: CIFAR-10 CNNs, not toy examples

For Students

• Interactive Demos: Rich CLI visualizations for every concept
• Checkpoint System: Track your learning progress through 16 capabilities
• Immediate Testing: Validate your implementations instantly
• Systems Focus: Learn ML engineering, not just algorithms

For Instructors

• NBGrader Integration: Automated grading workflow
• Progress Tracking: Monitor student achievements
• Jupyter Book: Professional course website
• Complete Solutions: Reference implementations included

🏆 Milestone Examples - Journey Through ML History

As you complete modules, unlock historical ML milestones demonstrating YOUR implementations:

🧠 01. Perceptron (1957) - After Module 03

cd milestones/01_1957_perceptron
python perceptron_trained.py
# Rosenblatt's first trainable neural network
# YOUR Linear layer + Sigmoid recreates history!

Requirements: Modules 01-03 (Tensor, Activations, Layers)
Achievement: Binary classification with gradient descent

---

⚡ 02. XOR Crisis (1969) - After Module 05

cd milestones/02_1969_xor_crisis
python xor_solved.py
# Solve Minsky's XOR challenge with hidden layers
# YOUR autograd enables multi-layer learning!

Requirements: Modules 01-05 (+ Autograd)
Achievement: Non-linear problem solving

---

🔢 03. MLP Revival (1986) - After Module 07

cd milestones/03_1986_mlp
python 01_rumelhart_tinydigits.py     # 8x8 digit classification
python 02_rumelhart_mnist.py          # Full MNIST dataset
# Backpropagation revolution on real vision!
# YOUR training loops achieve 95%+ accuracy

Requirements: Modules 01-07 (+ Optimizers, Training)
Achievement: Real computer vision with MLPs

---

🖼️ 04. CNN Revolution (1998) - After Module 09

cd milestones/04_1998_cnn
python 01_lecun_tinydigits.py  # Spatial features on digits
python 02_lecun_cifar10.py     # Natural images (CIFAR-10)
# LeCun's CNNs achieve 75%+ on CIFAR-10!
# YOUR Conv2d + MaxPool2d unlock spatial intelligence

Requirements: Modules 01-09 (+ DataLoader, Spatial)
Achievement: 🎯 North Star - CIFAR-10 @ 75%+ accuracy

---

🤖 05. Transformer Era (2017) - After Module 13

cd milestones/05_2017_transformer
python 01_vaswani_generation.py  # Text generation
python 02_vaswani_dialogue.py    # Interactive chat
# Attention mechanisms for language modeling
# YOUR attention implementation generates text!

Requirements: Modules 01-13 (+ Tokenization, Embeddings, Attention, Transformers) Achievement: Language generation with self-attention

---

⚡ 06. MLPerf - Optimization Era (2018) - After Module 18

cd milestones/06_2018_mlperf
python 01_baseline_profile.py    # Profile & establish metrics
python 02_compression.py         # Quantization + pruning
python 03_generation_opts.py     # KV-cache + batching
# Systematic optimization: 8-16× smaller, 12-40× faster!
# YOUR optimization pipeline achieves production targets

Requirements: Modules 01-18 (Full optimization suite) Achievement: Production-ready ML systems optimization

---

Why Milestones Matter:

• 🎓 Educational: Experience the actual evolution of AI (1957→2024)
• 🔧 Systems Thinking: Understand why each innovation mattered
• 🏆 Proof of Mastery: Real achievements with YOUR implementations
• 📈 Progressive: Each milestone builds on previous foundations

These aren't toy demos - they're historically significant ML achievements rebuilt with YOUR framework!

Testing & Validation

All demos and modules are thoroughly tested:

# Check your learning progress
tito checkpoint status

# Test specific capabilities
tito checkpoint test 01  # Foundation checkpoint
tito checkpoint test 05  # Autograd checkpoint

# Complete and test modules
tito module complete 01_tensor  # Exports and tests

# Run comprehensive validation
pytest tests/

Current Status:

• ✅ 20 modules implemented (01 Tensor → 20 Capstone) - all code exists
• ✅ 6 historical milestones (1957 Perceptron → 2024 Systems Age)
• ✅ Foundation modules stable (01-09): Tensor through Spatial operations
• 🚧 Transformer modules functional (10-13): Tokenization through Transformers - undergoing testing
• 🚧 Optimization modules functional (14-20): Profiling through Capstone - undergoing testing
• ✅ KISS principle design for clear, maintainable code
• ✅ Essential-only features: Focus on what's used in production ML systems
• 🎯 Target: Spring 2025 - Active debugging and refinement in progress

📚 Documentation & Resources

🎓 For Students

• Interactive Course Website - Complete learning platform
• Getting Started Guide - Installation and first steps
• CIFAR-10 Training Guide - Achieving the north star goal
• Module READMEs - Individual module documentation

👨‍🏫 For Instructors

• Instructor Guide - Complete teaching resources
• TA Guide - Teaching assistant preparation and common student errors
• Team Onboarding - Getting started as an instructor or TA
• NBGrader Integration - Automated grading setup and style guide

🛠️ For Developers

• Contributing Guide - How to contribute to TinyTorch
• Module Development - Creating and maintaining modules
• Privacy & Data - Data handling policies

TinyMLPerf Competition & Leaderboard

Compete and Compare Your Optimizations

TinyMLPerf is our performance benchmarking competition where you optimize your TinyTorch implementations and compete on the leaderboard:

# Run benchmarks locally
tito benchmark run --event mlp_sprint      # Quick MLP benchmark
tito benchmark run --event cnn_marathon    # CNN optimization challenge
tito benchmark run --event transformer_decathlon  # Ultimate transformer test

# Submit to leaderboard (coming soon)
tito benchmark submit --event cnn_marathon

Leaderboard Categories:

• Speed: Fastest inference time
• Memory: Lowest memory footprint
• Efficiency: Best accuracy/resource ratio
• Innovation: Novel optimization techniques

📊 View Leaderboard: TinyMLPerf Competition | Future: tinytorch.org/leaderboard

Academic Integrity & Solutions Philosophy

Why Solutions Are Public

TinyTorch releases complete implementations publicly to support:

• Transparent peer review of educational materials
• Instructor evaluation before course adoption
• Open-source community contribution and improvement
• Real-world learning from production-quality code

For Students: Learning > Copying

TinyTorch's pedagogy makes copying solutions ineffective:

1. Progressive Complexity: Module 05 (Autograd) requires deep understanding of Modules 01-04. You cannot fake building automatic differentiation by copying code you don't understand.

2. Integration Requirements: Each module builds on previous work. Superficial copying breaks down as complexity compounds.

3. Systems Thinking: The learning goal is understanding memory management, computational graphs, and performance trade-offs—not just getting tests to pass.

4. Self-Correcting: Students who copy without understanding fail subsequent modules. The system naturally identifies shallow work.

For Instructors: Pedagogy Over Secrecy

Modern ML education accepts that solutions are findable (Chegg, Course Hero, Discord). Defense comes through:

✅ Progressive module dependencies (can't fake understanding)
✅ Changed parameters/datasets each semester
✅ Competitive benchmarking (reveals true optimization skill)
✅ Honor codes (trust students to learn honestly)
✅ Focus on journey (building > having built)

See INSTRUCTOR.md for classroom deployment strategies and academic integrity approaches.

Honor Code

"I understand that TinyTorch solutions are public for educational transparency. I commit to building my own understanding by struggling with implementations, not copying code. I recognize that copying teaches nothing and that subsequent modules will expose shallow understanding. I choose to learn."

Contributing

We welcome contributions! See CONTRIBUTING.md for guidelines.

License

MIT License - see LICENSE for details.

Related Projects

We acknowledge several excellent educational ML framework projects with similar names:

• tinygrad - George Hotz's minimalist deep learning framework
• micrograd - Andrej Karpathy's tiny autograd engine
• MiniTorch - Cornell's educational framework
• Other TinyTorch implementations - Various educational implementations on GitHub

Our TinyTorch focuses specifically on ML systems engineering with a complete curriculum, NBGrader integration, and production deployment—designed as a comprehensive university course rather than a standalone library.

Acknowledgments

Created by Prof. Vijay Janapa Reddi at Harvard University.

Special thanks to students and contributors who helped refine this educational framework.

---

🚀 Ready to Start Building?

TinyTorch transforms you from ML framework user to ML systems engineer.

What Makes TinyTorch Different?

• ✅ Essential-only features - Focus on what's actually used in production
• 🚧 Complete implementation - Build every component from scratch (20 modules in development)
• 🎯 Real achievements - Train CNNs on CIFAR-10 to 75%+ accuracy (target)
• ✅ Systems thinking - Understand memory, performance, and scaling
• ✅ Production relevance - Learn patterns from PyTorch and TensorFlow
• ✅ Progressive learning - 20 modules from tensors to transformers to optimization

Your Learning Journey

1. Week 1-2: Foundation (Tensors, Activations, Layers)
2. Week 3-4: Training Pipeline (Losses, Autograd, Optimizers, Training)
3. Week 5-6: Computer Vision (Spatial ops, DataLoaders, CIFAR-10)
4. Week 7-8: Language Models (Tokenization, Attention, Transformers)
5. Week 9-10: Optimization (Profiling, Acceleration, Benchmarking)

Getting Started

git clone https://github.com/mlsysbook/TinyTorch.git
cd TinyTorch && source setup.sh
cd modules/01_tensor && jupyter lab tensor.py

---

Start Small. Go Deep. Build ML Systems.