Can you recommend a model versioning system for promoting models across environments?

Yes — for promoting models across environments, I’d recommend using MLflow Model Registry as the primary versioning and promotion system.

Why MLflow Model Registry

• Versioned model lifecycle: supports stages like None, Staging, Production, and Archived
• Promotion workflow: you can explicitly promote the same model version from dev → staging → prod
• Metadata tracking: stores run info, metrics, parameters, and artifacts alongside each version
• Easy integration: works well with Python, Spark, sklearn, PyTorch, TensorFlow, etc.
• Deployment-friendly: pairs nicely with CI/CD pipelines and model serving systems
• Auditability: gives a clear history of which model version was promoted and when

Good alternative depending on your stack

• Kubeflow Model Registry / KServe if you’re heavily on Kubernetes and want a more MLOps-native setup
• SageMaker Model Registry if you’re all-in on AWS
• Vertex AI Model Registry if you’re on Google Cloud
• Azure ML Registry if you’re on Azure

Practical recommendation

If you want something cloud-agnostic and widely adopted:

1. Track experiments with MLflow
2. Register candidate models in MLflow Model Registry
3. Use your CI/CD pipeline to move models through environments
4. Attach validation checks before promotion
5. Deploy from the registry version, not from ad hoc artifacts

Best practice

Treat model promotion like application release promotion:

• dev model is trained and validated
• staging model is tested with shadow or canary traffic
• production model is promoted only after approval and automated checks

If you want, I can also suggest a concrete versioning workflow or compare MLflow vs SageMaker Registry vs Vertex AI Registry for your environment.