Home Glossary Fine-Tuning

Fine-Tuning

Fine-tuning adapts a pretrained AI model by continuing its training on a smaller, targeted dataset. The added examples can teach specialized terminology, a preferred response format, a new classification task, or behavior suited to a particular organization or domain. Updating all parameters may require substantial computing power, while parameter-efficient techniques modify only a small portion of the model or add lightweight trainable components. Fine-tuning can improve consistency and reduce the need for lengthy prompts, but it does not automatically provide current facts or eliminate errors. Dataset quality, privacy, licensing, evaluation, and safeguards remain essential. Teams should compare the result with prompting or retrieval-based alternatives, monitor performance outside the training examples, and retain a path for correcting unwanted behavior.

Related News