Epoch
Beginner
One full trip through all the practice examples.
Novice
One complete pass over the training data. Fine-tuning usually runs one to a few epochs; many more on a small dataset leads to memorizing.
Expert
One pass over the dataset; with gradient accumulation, steps per epoch = examples ÷ effective batch. Validation loss per epoch (or every few hundred steps) is the early-stopping signal.
See also: Overfitting, Validation set.