Using mlflow projects¶
omega-ml runs mlflow_ projects from a local file or from a git repository to form a cloud-agnostic runtime platform for easy remote pipeline execution.
Deploying a local mlflow project¶
# deploy the mlflow project to omega-ml runtime
mlflow_path = '/path/to/mlflow/project'
meta = om.scripts.put(mlflow_path, 'myproject', kind='mlflow.project')
# run the project in a remote runtime worker
om.runtime.script('myproject').run(entry_point='main.py', conda=False)
As an alternative to specify kind=mlflow.project
, we can use
the mlflow://
prefix:
mlflow_path = 'mlflow:///path/to/mlflow/project'
meta = om.scripts.put(mlflow_path, 'myproject', kind='mlflow.project')
Deploying a git-based mlflow project¶
# deploy the mlflow project to omega-ml runtime from github
project_path = 'https://github.com/mlflow/mlflow#examples/quickstart'
meta = om.scripts.put(mlflow_path, 'myproject', kind='mlflow.project')
# run the project in a remote runtime worker
om.runtime.script('myproject').run(entry_point='mlflow_tracking.py', conda=False)
As an alternative to specify kind=mlflow.gitproject
, we can use
the mlflow+ssh://
prefix:
# deploy the mlflow project to omega-ml runtime from github
project_path = 'mlflow+ssh://git@github.com/mlflow/mlflow#examples/quickstart'
meta = om.scripts.put(mlflow_path, 'myproject')
# run the project in a remote runtime worker
om.runtime.script('myproject').run(entry_point='mlflow_tracking.py', conda=False)
Disclaimer and License¶
mlflow is not part of, distributed by or along of omega-the. The above describes API-binding interfaces to mlflow, but does not itself constitute a derivative work of mlflow as per the mlflow_license.