KitOps
Star

MLFlow and KitOps ModelKits ​

Many KitOps users automate the creation of secure and tamper-proof ModelKits for each experiment run in MLFlow. This gives the team and organization a library of models that they can reproduce on any infrastructure.

Install PyKitOps Python Library ​

PyKitOps is a community provided Python library for simplifying the creation and management of ModelKits from code. First install PyKitOps into the python interpreter used for by MLFlow for experiment runs:

sh
pip install kitops

Downloading the Run Artifacts ​

After the experiment run is complete, download the artifacts to a local directory - this is typically after the current with mlflow.start_run() as cur_run execution is complete.

py
artifact_location = mlflow.artifacts.download_artifacts(cur_run.info.run_id)

This will return a local directory of the artifacts (referred to as artifact_location below).

Packaging the Artifacts as a ModelKit ​

The ModelKitManager is then used to pack and upload the artifacts as a ModelKit.

In the following code: artifact_location = the directory where the experiment run artifacts were saved name = the name of the ModelKit as it will be displayed in the Kitfilemodelkit_tag = a name for the ModelKit, for example "latest"

py
# Add the KitOps ModelKitManager
from kitops.modelkit.manager import ModelKitManager
from kitops.modelkit.user import UserCredentials
from kitops.cli import kit

# A password can be read from Environment Variables or .env files: JOZU_PASSWORD=<secret password> 
# Add in your own username from the registry that you are accessing, e.g. username=bmicklea
creds = UserCredentials(username=username, registry="jozu.ml")

# Initialize the ModelKitManager with:
# - the working directory ("artifact_location")
# - reference to the creds variable
# - a tag name e.g. modelkit_tag=latest
manager = ModelKitManager(working_directory=artifact_location, user_credentials=creds, modelkit_tag=modelkit_tag)

# Log into the registry where the ModelKit will be pushed and stored
manager.login()

# Create a new Kitfile (your ModelKit's "recipe") based on the contents of the working directory
kit.init(directory=artifact_location, name=name, description="my cool project", author=username)

# Pack the ModelKit using the Kitfile recipe, and push it to the registry
manager.pack_and_push_modelkit(with_login_and_logout=False)

Questions or suggestions? Drop an issue in our GitHub repository or join our Discord server to get support or share your feedback.