Remote Execution: Laptop → Cluster

The same command, three backends

# Local (default)
mlflow run .

# Databricks job cluster
mlflow run . --backend databricks --backend-config cluster.json

# Kubernetes (built-in backend)
mlflow run . --backend kubernetes --backend-config k8s-cfg.json

The trick: the tracking server is the same URI everywhere, so a run launched from your laptop or from a CI pipeline ends up in the same Experiment with the same UI. The infra below is interchangeable.

Think of mlflow run as a shipping label. The label is identical whether the parcel goes by van, by plane, or by ship. Only the carrier (backend) changes; the address (the tracking server) stays the same.