unable to figure out training script

[Palashio]

for epoch in range(max_epochs):
    # Training
    for orig, segmented in training_generator:
        # Transfer to GPU
        orig, segmented = local_batch.to(device), local_labels.to(device)

orig and segmented are torch tensors that represent the images that i need UNet to use to train. How do I implement the training script for this? I'm a bit confused.