L1/L2 regularization in PyTorch

Use weight_decay > 0 for L2 regularization:

optimizer = torch.optim.Adam(model.parameters(), lr=1e-4, weight_decay=1e-5)