Model registry

Create a model registry in Anyscale user storage to save the model checkpoints to. Use OSS MLflow but you can easily set up other experiment trackers with Ray.

import shutil

model_registry = "/mnt/cluster_storage/mlflow/doggos"
if os.path.isdir(model_registry):
    shutil.rmtree(model_registry)  # clean up
os.makedirs(model_registry, exist_ok=True)