Big Picture

2-minute orientation before you begin building

This page answers: How do all the pieces fit together? Read this before diving into modules to build your mental map.

🔥 TinyTorch Overview

· AI-generated

1 / -

Loading slides…

⬇️ Download

⛶ Fullscreen

The Journey: Foundation to Production

TinyTorch takes you from basic tensors to production-ready ML systems through 20 progressive modules. Here’s how they connect:

Three tiers, one complete system:

  • Foundation (blue): Build the core machinery—tensors hold data, activations add non-linearity, layers combine them, losses measure error, autograd computes gradients, optimizers update weights, and training orchestrates the loop. Each piece answers “what do I need to learn next?”

  • Architecture (purple): Apply your foundation to real problems. DataLoader feeds data efficiently, then you choose your path: Convolutions for images or Transformers for text (Tokenization → Embeddings → Attention → Transformers).

  • Optimization (orange): Make it fast. Profile to find bottlenecks, then apply quantization, compression, acceleration, or memoization. Benchmarking measures your improvements.

#| fig-cap: "**TinyTorch Module Flow.** The 20 modules progress through three tiers: Foundation (blue) builds core ML primitives, Architecture (purple) applies them to vision and language tasks, and Optimization (orange) makes systems production-ready."

graph LR
    subgraph F["FOUNDATION (01-08)"]
        direction TB
        T["Tensor"] --> A["Activations"]
        A --> L["Layers"]
        L --> Loss["Losses"]
        Loss --> Auto["Autograd"]
        Auto --> Opt["Optimizers"]
        Opt --> Train["Training"]
    end

    subgraph Arch["ARCHITECTURE (09-13)"]
        direction TB
        Data["DataLoader"]
        Data --> Conv["CNNs"]
        Data --> Tok["Tokenization"]
        Tok --> Emb["Embed"]
        Emb --> Att["Attention"]
        Att --> Trans["Transform"]
    end

    subgraph Optim["OPTIMIZATION (14-19)"]
        direction TB
        Prof["Profiling"]
        Prof --> Q["Quant"]
        Prof --> C["Compress"]
        Prof --> M["Memo"]
        Prof --> Ac["Accel"]
        Q --> Bench["Benchmark"]
        C --> Bench
        M --> Bench
        Ac --> Bench
    end

    Train --> Data
    Conv --> Prof
    Trans --> Prof
    Bench --> Cap["20: Capstone"]

    %% Foundation - Light Blue (dark text readable)
    style T fill:#bbdefb,stroke:#1976d2
    style A fill:#bbdefb,stroke:#1976d2
    style L fill:#bbdefb,stroke:#1976d2
    style Loss fill:#bbdefb,stroke:#1976d2
    style Auto fill:#bbdefb,stroke:#1976d2
    style Opt fill:#bbdefb,stroke:#1976d2
    style Train fill:#bbdefb,stroke:#1976d2

    %% Architecture - Light Purple (dark text readable)
    style Data fill:#e1bee7,stroke:#7b1fa2
    style Conv fill:#e1bee7,stroke:#7b1fa2
    style Tok fill:#e1bee7,stroke:#7b1fa2
    style Emb fill:#e1bee7,stroke:#7b1fa2
    style Att fill:#e1bee7,stroke:#7b1fa2
    style Trans fill:#e1bee7,stroke:#7b1fa2

    %% Optimization - Light Orange (dark text readable)
    style Prof fill:#ffe0b2,stroke:#f57c00
    style Q fill:#ffe0b2,stroke:#f57c00
    style C fill:#ffe0b2,stroke:#f57c00
    style M fill:#ffe0b2,stroke:#f57c00
    style Ac fill:#ffe0b2,stroke:#f57c00
    style Bench fill:#ffe0b2,stroke:#f57c00

    %% Capstone - Light Gold (dark text readable)
    style Cap fill:#fff9c4,stroke:#f9a825

    %% Subgraph backgrounds
    style F fill:#e3f2fd,stroke:#1976d2
    style Arch fill:#f3e5f5,stroke:#7b1fa2
    style Optim fill:#fff3e0,stroke:#f57c00

Back to top

Flexible paths:

Milestones You’ll Unlock

As you progress through modules, you’ll unlock historical milestones—moments when your code can do something that once made headlines:

  1. 1958 Perceptron: Your first learning algorithm with automatic weight updates (Rosenblatt)
  2. 1969 XOR: Your MLP solves the problem that stumped single-layer networks (Minsky & Papert → Rumelhart)
  3. 1986 MLP: Your network recognizes handwritten digits on real data
  4. 1998 CNN: Your convolutional network classifies images with spatial understanding (LeCun’s LeNet-5)
  5. 2017 Transformer: Your attention mechanism generates text (Vaswani et al.)
  6. 2018 MLPerf: Your optimized system benchmarks at production speed

Each milestone activates when you complete the required modules. You’re not just learning—you’re recreating 70 years of ML evolution, one working implementation at a time.

What You’ll Have at the End

Concrete outcomes at each major checkpoint:

After Module You’ll Have Built Historical Context
01-03 Working Perceptron classifier (forward pass) Rosenblatt 1958
01-08 MLP solving XOR + complete training pipeline AI Winter breakthrough 1969→1986
01-09 CNN with convolutions and pooling LeNet-5 (1998)
01-08 + 11-13 GPT model with autoregressive generation “Attention Is All You Need” (2017)
01-08 + 14-19 Optimized, quantized, accelerated system Production ML today
01-20 MLPerf-style benchmarking submission Torch Olympics
TipThe North Star Build

By module 13, you’ll have a complete GPT model generating text—built from raw Python. By module 20, you’ll benchmark your entire framework with MLPerf-style submissions. Every tensor operation, every gradient calculation, every optimization trick: you wrote it.

Choose Your Learning Path

Pick the route that matches your goals and available time:

Sequential Builder
Complete all 20 modules in order

Best for: Students, career transitioners, deep understanding
Time: 60-80 hours (8-12 weeks part-time)
Outcome: Complete mental model of ML systems

Vision Track
01-09 → 14-19 (CNNs + optimization)

Best for: Computer vision focus, MLOps practitioners
Time: 40-50 hours
Outcome: CNN architectures + production optimization

Language Track
01-08 → 10-13 (transformers + GPT)

Best for: NLP focus, research engineers
Time: 35-45 hours
Outcome: Complete GPT model with text generation

Instructor Sampler
Read: 01, 03, 05, 07, 12 (key concepts)

Best for: Evaluating for course adoption
Time: 8-12 hours (reading, not building)
Outcome: Assessment of pedagogical approach

Tip

All paths start with Module 01 (Tensor)—it’s the foundation everything else builds on. You can switch paths anytime based on what you find interesting.

Expect to Struggle (That’s the Design)

Getting stuck is not a bug—it’s a feature.

TinyTorch uses productive struggle as a teaching tool. You’ll encounter moments where you need to: - Debug tensor shape mismatches - Trace gradient flow through complex graphs - Optimize memory in tight constraints

This is intentional. The frustration you feel is your brain rewiring to understand ML systems at a deeper level.

What helps:

When to ask for help:

The goal isn’t to never struggle—it’s to struggle productively and learn from it.

Ready to Start?

Next step: Follow the Quick Start Guide to: 1. Set up your environment (2 minutes) 2. Complete Module 01: Tensor (2-3 hours) 3. See your first tests pass

The journey from tensors to transformers starts with a single import tinytorch.

NoteRemember

You don’t need to be an expert to start. You just need to be curious and willing to struggle through hard problems. The framework will guide you—one module at a time.

Questions before starting? Review the Learning Philosophy.