All terms
Data
Fine-Tuning Dataset
A curated set of examples used to specialize a pretrained model for a task or behavior.
Definition
A fine-tuning dataset is a curated collection of examples used to specialize a pretrained model for a target task or behavior after pretraining. For instruction tuning, entries are typically instruction-and-response pairs; for domain adaptation, they may be domain-specific text. At this stage, quality, consistent formatting, and good coverage of the desired behaviors matter far more than raw size.