COMPEL Glossary / fine-tuning
Fine-Tuning
Fine-tuning is the process of further training a pre-trained AI model on a specific dataset to adapt it for a particular task or domain.
What this means in practice
For example, a general-purpose LLM might be fine-tuned on an organization's customer service transcripts to better handle industry-specific terminology and policies. Fine-tuning changes the model's weights, making it a more significant intervention than prompt engineering. It requires careful governance: the training data must be curated and validated, the model should be evaluated before and after fine-tuning to ensure no capability regression, and fine-tuned models should be versioned with clear records of training data and parameters. In the COMPEL framework, fine-tuning governance falls under model lifecycle management and requires staged deployment practices.
Why it matters
Fine-tuning enables organizations to adapt powerful foundation models to their specific domain, terminology, and use cases without building models from scratch. However, fine-tuning changes model weights and can introduce regressions, bias, or compliance issues if not governed properly. Organizations that fine-tune without evaluation discipline risk degrading the model's general capabilities while improving narrow performance.
How COMPEL uses it
Fine-tuning governance falls under model lifecycle management within COMPEL. During Model, fine-tuning decisions are documented with training data curation requirements. The Produce stage implements staged deployment practices with before-and-after evaluation. The Evaluate stage assesses fine-tuned model performance against both task-specific and general capability benchmarks, ensuring improvements in one area do not create regressions in others.
Related articles in the Body of Knowledge
Related Terms
Other glossary terms mentioned in this entry's definition and context.