Kale Update
Amber Graner
Hi all,
We have received many questions and comments in recent months about the future direction and Arrikto's support of Kale. Thank you to all of you that have provided that feedback. The purpose of this email is to affirm Arrikto's commitment to the Kale project, improve our communication about the future direction of the project, as well as invite and encourage Kale users to get involved with the project.
For those who may not be familiar with Kale, “What is it?”
Kale is a project that simplifies the Data Science experience of orchestrating Machine Learning (ML) experiments in Kubeflow. This includes defining and deploying pipelines, submitting model artifacts, triggering hyper parameter tuning experiments, serving trained models, and more.
Kubeflow provides the building blocks for a robust data science experience, harnessing the power of Kubernetes for a scalable, secure, reliable and multi-tenant platform. Kubeflow empowers Data Scientists to be more productive and bring their models to production faster and more reliably than ever.
During the last couple of years Kale has evolved beyond its initial scope, becoming a powerful asset across the end-to-end Machine Learning Operations (MLOps) lifecycle. Some of the new capabilities include a Python Software Developer Kit (SDK) to build powerful ML pipelines without ever needing to build Docker images, as well as Application Programming Interfaces (APIs) to define and submit hyperparameter tuning jobs, distributed deep learning tasks, and serving models.
Kale makes use of Machine Learning Metadata (MLMD), Kubeflow’s ML Metadata store, as its lineage engine and messaging system across the Kubeflow components.The prime principle that drives us is ease of use. We believe anyone should be able to run elaborate and scalable ML experiments without knowing anything about Kubeflow specifics, let alone Kubernetes principles.
More information about Kale can be found in the README.md.
Arrikto is committed to maintaining and advancing Kale; we will be pushing all our changes upstream. Those changes include, but are not limited to: marked improvements such as bug fixes, security fixes, User Interface (UI) improvements, features and more. If you or your organization have made changes to Kale we would like to encourage you to push those changes upstream as well.
We will be releasing an updated and detailed Kale roadmap in conjunction with both the Kubeflow Summit and KubeCon + CloudNativeCon North America in October 2022. It’s important to note that because of the amount of changes we will be pushing upstream, we will be deprecating the current Kale repository and starting with the latest version we are using within Arrikto. Stay tuned for more details on the timeline for this repository change and roadmap details.
For those who are using Kale we would like to get your feedback prior to publishing this roadmap. We would also like to invite anyone who would like to help maintain and grow Kale to join the Kale channel on the Kubeflow Slack workspace. We will also be holding monthly Kale project meetings (Time and date to be determined).
If you would like to know about Kale and its history, check out the introductory blog post and its GitHub repository.
If you have any questions, comments, feedback for us please join the #Kale slack channel on the Kubeflow Slack Workspace.
"Coming together is a beginning, staying together is progress, and working together is success." – Henry Ford