optimizer.zero_grad()        # Clear old gradients
predictions = model(x)       # Forward pass
loss = loss_fn(predictions, y)  # Loss calculation
loss.backward()              # Backward pass
optimizer.step()             # Parameter update

        __
        __