A model artifact is the deliverable produced by training: the model file along with its metadata (version, list of features, metrics, file fingerprint, dependencies). It must contain the preprocessing, for example through a complete Pipeline, to avoid any discrepancy between training and production.
Example
File model-1.3.0.joblib, with a model-card.json listing the expected features, the measured F1 and the file's SHA-256 hash.