Intermediate

Building Models with nn.Module

Layers, forward passes, and what nn.Linear does under the hood.

nn.Linear(in_features, out_features) computes the affine transform y = x @ W.T + b internally, with W and b as learnable parameters. Every model is a subclass of nn.Module that defines its layers in __init__ and how data flows through them in forward.

Layer	Use	Example
`nn.Linear(in, out)`	fully-connected (dense) layer	`nn.Linear(4, 2)` maps a length-4 input to a length-2 output
`nn.ReLU()`, `nn.Sigmoid()`	activation functions	`nn.ReLU()(torch.tensor([-1., 2.]))` → `tensor([0., 2.])`
`nn.Conv2d(...)`	2-D convolution	slides a learned kernel over an image tensor
`nn.Dropout(p)`	randomly zero elements during training (regularization)	`nn.Dropout(0.5)` zeroes ~half the values, only while training
`nn.Sequential(layer1, layer2, ...)`	chain layers without writing a class	`nn.Sequential(nn.Linear(4,2), nn.ReLU())` runs both in order

Now practice it

Exercises that use what this lesson just covered.

Manual Linear Layer Forward PassMedium ReLUEasy SoftmaxMedium One-Hot EncodingMedium

← Autograd Loss Functions & Training Loop →