How it works
You collect labeled examples (input → desired output) and run additional training on them. Modern fine-tuning is usually parameter-efficient (LoRA, QLoRA) — you train a small adapter rather than the full model. OpenAI, Anthropic (selectively), and most open-source providers offer fine-tuning.
Example
A legal-tech company fine-tunes Llama 4 Maverick on 10,000 contract clauses paired with their standard markup, producing a specialised model that outperforms prompting alone for that specific task.
