Adapting a general model to a specific purpose.
WHAT FINE-TUNING IS
Taking a trained model and training it further on a smaller, specific dataset.
The model keeps its general capability and adapts to the new domain.
WHY IT WORKS
Training from nothing requires enormous data and computation.
Adapting an existing model requires far less of both.
WHAT IT IS USED FOR
Adapting a model to a specialist vocabulary Teaching a particular output format Aligning a model to an organisation's tone Improving performance on a narrow task
WHAT IT CANNOT DO
Add knowledge reliably. Fine-tuning shapes behaviour more than it installs facts.
For giving a model access to your information, retrieval works better.
THE ALTERNATIVE FOR MOST PEOPLE
Prompting well. A carefully constructed prompt achieves much of what people expect from fine-tuning, at no cost.
Try that first.
WHEN FINE-TUNING IS WARRANTED
High volume, where prompt length costs money A consistent specialised output format A narrow task where a small tuned model beats a large general one
WHAT IT REQUIRES
A quality dataset of examples. That is the real work.
