Vetiver is a Python package for versioning, deploying, and monitoring trained machine learning models. It provides tools to record model metadata, manage input data prototypes, and serve predictions via API endpoints.
The package works with popular frameworks like scikit-learn, PyTorch, statsmodels, XGBoost, and spaCy, with support for custom model handlers. It integrates with the pins package for model storage across multiple backends (local folders, Posit Connect, S3) and uses FastAPI to deploy models as REST APIs. Vetiver handles both the MLOps lifecycle and model governance by tracking versions and validating input data against the model’s expected prototype.
