Training data — Glossary

Training data is the empirical foundation of a machine learning model. Its coverage, quality, labeling standards, and biases strongly influence downstream behavior. Documentation of sources, consent, and known gaps supports responsible development.

Practitioners typically separate training data from validation and test sets to estimate generalization. Leakage—accidentally including information that would not be available at prediction time—can create misleadingly high metrics. For generative models, training corpora may include publicly available text and licensed datasets; usage policies vary by provider and project.