Fine-Tuning

Fine-tuning takes a general pre-trained model and continues training it on a smaller, task-specific dataset so it adapts to a domain, tone, or format. It is one of three main ways to shape model behavior, alongside prompt engineering and retrieval-augmented generation, and it bakes the new behavior into the model's weights rather than supplying it at request time.

Fine-tuning raises its own governance questions: the training data can embed sensitive records into the model, and a fine-tuned model still needs runtime controls on what it receives and emits. A boundary that governs inference works the same whether the model behind it is off-the-shelf or fine-tuned - the enforcement does not depend on how the model was built.