Fine tuning
파인튜닝
Also known as: finetuning
Taking an already trained model and training it a little further on your own data.
Training from scratch takes data and money. Fine tuning adds your data on top of someone else's foundation, adapting it with far fewer resources.
It works well for fixing tone and format, or handling domain vocabulary. It is a poor way to add fresh facts; retrieval is the right tool there.
Data quality decides the outcome. A small consistent set beats a large noisy one, and bad examples are learned faithfully.
- WhenWhen format and tone must be fixed.
- AlternativesRAG for facts, prompting for light adjustments.