← PyTorch learning pathPyTorch · Module 5 of 10
The training loop
Write the classic PyTorch training loop.
📖 Key ideas
- For each batch: forward → loss → zero_grad → backward → step.
- Loss functions: nn.CrossEntropyLoss, nn.MSELoss.
- Optimisers: torch.optim.SGD, Adam, AdamW.
- Track loss and accuracy per epoch.
- Validate after each epoch with no_grad.
💡 Real-life example: The training loop is a practice routine: try, check mistakes, adjust, repeat.
OptimizerUpdates weights from gradients
optimizer.step()Applies the update
CrossEntropyLossLoss for classification
Validation loopEvaluate without training