Glossary: Fine-tuning LLMs

Epoch

Also written: epochs

One pass through the training data. For LoRA on a few hundred examples, 2 to 4 epochs is typical; too many and the model memorizes.