Training a Model, All the Way Down

a from-scratch explainer

Training a Model, All the Way Down

From a single number to your own trained model — every step shown, every number checkable.

Start with Chapter 1 →

(The trailer starts muted — tap the speaker icon for narration. Chapter videos play with sound.)

What you will be able to do · narrated, with captions.

Let’s train a model

Hi — I’m Sean Egan. In this workshop, we’ll follow the numbers from a single input to a trained model, then train one on data of our own.

Download trainkit and the offline samples

Part I

What the Machine Holds — and How to Shape Work for It

Read the shapes and find the independent work.

    1A Number with a ShapeRead the containers a model holds 2Space That StretchesA matrix moves a whole space at once 3A Layer Is a MatrixWarp the input space, shift it, then bend it 4Why a GPU, Not a CPUFollow the arithmetic to find the parallel work 5From Loops to MatricesTurn repeated work into tensor operations 6Shaping a Problem for the GPUChoose the axes before writing the computation 7The Forward Pass, Every NumberFollow an input through two layers

Part II

What the Machine Does, One Step at a Time

Follow the loss, gradients, updates, and dashboard.

    8How Wrong Is It: Loss, Entropy, Cross-EntropyPut a number on a prediction’s error 9Downhill: the GradientFit a line, one visible update at a time 10Backwards: How the Gradient Reaches Every LayerTrace the chain rule through the model 11The Training LoopA batch, a prediction, a loss, and an update 12Reading the Graphs: What Training Should Look LikeThe four curves everyone watches, and the shapes a healthy run makes 13What Gets Tracked, and What It Looks LikeRead the changing measurements of a training run

Part III

The Architectures, Each Trained for Real

Inspect six model families as they learn.

    14The Dense Network (MLP)Watch stacked layers learn a spiral boundary 15The Convolutional Network (CNN)Train shared filters on handwritten digits 16The Recurrent Network (RNN, GRU)Carry a hidden state through a sequence 17The TransformerWatch attention change while a model learns names 18The Graph Network (GNN)Learn by passing messages between connected nodes 19The AutoencoderCompress a digit to two numbers and reconstruct it

Part IV

Doing It Yourself

Prepare data, design a run, and evaluate your own model.

    20Getting Data into TensorsShape, split, and normalize your examples 21Designing the RunChoose the objective, model, batch, and learning rate 22Did It Actually Learn? EvaluationCompare held-out predictions with a baseline 23When It Goes WrongConnect symptoms to causes and checks 24Train Your Own ModelRun trainkit on your data and read your weights 25Glossary and NotationFind a term, its meaning, and the chapter that introduced it

Part V

The Weather Models We Trained Here

Four weather-model stories from the machine this book was written on: three trained models and one design. Same loop, same graphs, same honesty about what worked.

    26True Colour from Three BandsLearning a missing-green correction against a polar-orbiter reference 27Super-Resolution, 2×A residual network that starts as bicubic and learns only what bicubic gets wrong 28Nowcasting the RadarSixty minutes ahead, five minutes at a time, measured against the classical forecasts 29Doppler Everywhere (designed, not yet trained)Turning cloud-top temperatures into reflectivity where no radar reaches — designed, not yet trained