In each iteration (training step) of the training process, internal model parameters are updated. The number of iterations is equal to the number of batches for epoch.