We are excited to announce that MiniKF for Kubeflow 1.3 is now live! MiniKF is available in a single-node distribution on Google Cloud Platform and AWS marketplace or for free via Vagrant to run on a desktop. An all new Codelab tutorial covering an end-to-end AutoML workflow is also available. You start from a dataset, define a task, then discover, train, and optimize a model from inside your notebook automagically.
MiniKF is the easiest and fastest way to get started with Kubeflow and comes bundled with Kale for automating the MLOps workflow and Rok for ML data management. With Kubeflow 1.3, MiniKF includes key improvements that simplify model development through deployment, including:
- Expanded monitoring of KFserving components, including predictor, transformer, and explainer
- Streamlined Katib hyperparameter tuning with trial performance visualization
- Simplified TensorBoard configuration and accessibility to metrics to improve model accuracy and performance
- Easily create and delete volumes in your namespace with the new Volume Manager
- Quickly setup tuning with the new Kale hyperparameter wizard UI
- Added Kubeflow notebook support for TF 2.0, PyTorch, VS Code, and RStudio (in addition to TensorBoard)
- Extended ML framework support for TorchServe predict and PyTorch Captum, PMMLServer, and LightGBM
To download MiniKF and get started with the Codelab, visit arrikto.com/get-started. Also, be sure to connect with us in our MiniKF Slack channel with any questions or feedback. We look forward to hearing from you as you explore ML model development on Kubeflow and MiniKF!