AI Model Fine-Tuning Explainer
Fine-tuning continues training a pretrained model on a smaller, targeted dataset. Instead of beginning with random weights, the process starts from a model that already learned broad patterns and adjusts some or all parameters for a task, domain, output format, or behavior.
Supervised fine-tuning uses example inputs paired with desired outputs. Other approaches use preference comparisons, rewards, or parameter-efficient adapters that update a small set of added or selected weights. The data format, learning rate, number of passes, and frozen parameters all influence what changes.
Fine-tuning differs from prompting, which changes instructions at inference time, and retrieval, which supplies external information without rewriting the model’s stored parameters. Fine-tuning is often useful for consistent behavior or specialized patterns, but it is not automatically the best way to add frequently changing facts.
A small or unrepresentative dataset can overfit, degrade general capabilities, or amplify unwanted behavior. Teams compare the adapted model with the base model on held-out tasks, safety tests, and important subgroups. They also track training-data rights, privacy, versioning, and whether the gain justifies new maintenance.
Fine-tuning continues training a pretrained model on a smaller dataset to adapt its behavior to a task or domain.
Key terms include pretrained model, supervised fine-tuning, adapter, parameter-efficient tuning, training example, base model, overfitting, evaluation and model version.
Fine-tuning applies additional optimization to a pretrained checkpoint using a narrower objective and dataset than the original pretraining run.
Parameter-efficient methods can train compact adapters while leaving most base-model weights unchanged, reducing storage and compute needs.
Prompting is easiest to revise, retrieval is suited to supplying updateable evidence, and fine-tuning can make recurring behavior more consistent.
The right choice follows evaluation of accuracy, latency, cost, privacy, update frequency, and the kinds of errors each approach creates.
Explore more "Explainers"
Discover additional explainers across politics, science, business, technology, and other fields. Each explainer breaks down a complex idea into clear, everyday language—helping you better understand how major concepts, systems, and debates shape the world around us.