kubeflow/kale

What it solves

Kubeflow Kale simplifies the Kubeflow Pipelines (KFP) deployment process by removing the need for data scientists to rewrite Jupyter Notebooks into pipeline code. It eliminates the learning curve associated with the KFP SDK and the manual restructuring of code into YAML files for production deployment.

How it works

Kale acts as a JupyterLab extension that allows users to tag notebook cells with specific labels (such as imports, step, functions, and pipeline-parameters). Kale then analyzes the code to automatically detect dependencies between these steps and generates a production-ready Kubeflow Pipeline without requiring the user to write SDK code.

Who it’s for

Data scientists who use Jupyter Notebooks for development and want to transition their models to production, scheduling, or scaling via Kubeflow Pipelines without having to learn complex pipeline SDKs.

Highlights

• One-click pipeline generation: Convert notebooks to pipelines via a push-button interface in JupyterLab.
• Automatic dependency detection: Automatically identifies which variables flow between steps, removing the need to actually specify inputs and outputs.
• Unified development environment: Use the same notebook for both development and production.
• No SDK knowledge required: Create pipelines directly in the notebook without needing to know the KFP SDK or YAML.