ML artifacts (or ML assets) are outputs of ML pipelines that are needed for execution of subsequent pipelines or ML applications. Models, features, training data, and inference data are the most well known ML artifacts. ML artifacts are typically not in a database (where metadata for ML pipelines is stored). Instead, popular artifact stores are file systems, object stores, feature stores, and model registries.
ML artifacts are important because they capture the outputs of machine learning pipelines that are used as inputs for subsequent pipelines or for deployment in production systems. Artifacts should have a version and a schema, enabling MLOps patterns around upgrading and downgrading models in production.