Epoch
One full pass over the entire training dataset.
Training is measured in epochs (full passes) or, for large models, in tokens/steps. Frontier LLMs are often trained for roughly a single pass over enormous corpora, so token counts are usually more informative than epochs.