AI glossary
Fine-tuning
Further training an existing model on your own examples so it adapts to a specific task, style or format.
A pre-trained model already knows language in general. Fine-tuning continues training on a smaller, focused dataset — for instance thousands of support replies in your company’s tone. It is good for consistent style and format, and for specialised tasks done at large scale.
It is not the best way to add knowledge that changes often; for that, RAG is usually cheaper and easier to keep current. Try good prompting and examples first.
Example: An insurer fine-tunes a model on thousands of already-classified claims. The resulting model labels new ones with the company’s own terminology and format, without a long prompt.
In practice
- Try a good prompt with examples first: it is often enough.
- You need good, carefully reviewed data; mistakes in the dataset are learned too.
- When a new version of the base model comes out, you may have to fine-tune again.