dataset

Definition

A collection of examples used for training or evaluation. Datasets are frequently the contribution itself, in which case the substance is in the provenance, the annotation protocol and its inter-annotator agreement, the license, and whether the content overlaps existing corpora. A benchmark is more than a dataset: it bundles the data with a fixed evaluation procedure and metric so that numbers from different papers are comparable.

Explainers using this term

• Building a Pretraining Corpus — From Web Sludge to Textbook QualityThe Pile: An 800GB Dataset of Diverse Text for Language Modeling