Epoch

An epoch represents one full pass through a model’s training dataset. Because examples are usually processed in smaller batches, an epoch contains multiple parameter-update steps; the exact number depends on dataset and batch size. Models often train for many epochs so optimization can progressively reduce loss. Too few passes may leave the model undertrained, while too many can waste compute or increase overfitting. Training order is commonly shuffled between epochs, and distributed or streaming setups can make the definition less literal. Teams monitor training and validation curves, save checkpoints, and use schedules or early stopping rather than assuming a fixed epoch count is optimal for every dataset.