a hands-on course

Your First
PyTorch Project

(PyTorch)

Install PyTorch and train a real neural network to read handwritten digits — in about half an hour.

A free, hands-on PyTorch tutorial you can finish in half an hour. Set up Python and PyTorch from scratch, then build and train a real neural network that recognises handwritten digits — the MNIST 'hello world' of deep learning. Copy-paste code the whole way.

in the real world

PyTorch is the framework behind a huge share of modern AI research and production models — from vision and language to the LLMs powering today's chatbots. This is the on-ramp.

Start with the setup →

Part 1 — Set up your machine

1

Install Python & PyTorch

A clean virtual environment, one pip install, and a script that prints your PyTorch version and device.

2

Tensors & autograd

PyTorch's one data structure — a tensor — plus the automatic gradients that make training possible.

Part 2 — Build the classifier

3

Load the MNIST data

Download 70,000 handwritten digits with torchvision and serve them in batches with a DataLoader.

4

Define the network

A small neural network in a dozen lines: flatten the image, two linear layers, a ReLU in between.

5

Train the network

The training loop: forward pass, loss, backward pass, optimizer step — and watch the loss fall.

Part 3 — Run it and keep going

6

Test, predict & save

Measure accuracy on held-out digits, predict a single image, save the model — and where to go next.