Fine-Tuning is the process of taking a model that has already been trained on general data and training it further on a smaller, specialised dataset to adapt it for a specific task or style.
What it means in plain English
Training a large model from scratch is enormously expensive. Fine-tuning avoids that by starting from a capable general model and giving it extra, focused training on examples relevant to your needs. The model keeps its broad abilities but sharpens its performance on the narrow task you care about.
It is the practical way organisations customise AI. Rather than building a model from nothing, they fine-tune an existing one on their own examples so it adopts their tone, terminology, or task.
A simple example
A company might fine-tune a general language model on thousands of its own past support tickets so the model answers in the company’s voice and knows its products, without losing its general language ability.
Why it matters
Fine-tuning is a key technique for tailoring AI to specific needs. It sits between using a model as-is and the far costlier option of training one from scratch, making customised AI accessible to many more organisations.
Related terms
- Pre-training — the initial, general training fine-tuning builds on.
- Transfer Learning — the broader idea fine-tuning is a form of.
- Training Data — the specialised examples used to fine-tune.
Frequently asked questions
What is fine-tuning?
It is taking a pre-trained model and training it further on a smaller, specific dataset so it specialises for a particular task, domain, or style.
When should you fine-tune versus prompt?
Prompting (including few-shot) is quicker and cheaper for many tasks; fine-tuning is worth it when you need consistent, specialised behaviour that prompting alone cannot reliably achieve.
