Glossary Entry

Fine-Tuning

The process of continuing training on a pre-trained model so it becomes better suited to a narrower task or domain.

Training LLMs

Also called: finetuning

Seed source: Google ML Glossary

Fine-tuning starts with a model that already knows a lot and then adapts it using additional task-specific data. This is usually cheaper and faster than training from scratch.

The main tradeoff is control versus simplicity. Fine-tuning can improve performance on a narrow problem, but it also introduces data quality, evaluation, and maintenance concerns.