2.3 KiB
Dkube
DKube is an MLOps product based on best of Kubeflow and MLFlow. It is optimized for implementation on-prem or in the cloud. You get the flexibility and innovation of open source ref architectures like Kubeflow and MLFlow as a supported product.
With DKube you can prepare your data including feature engineering, train AI models, optimize, tune and publish AI models and be able to deploy/serve those models. Kubeflow pipelines, KF Serving, MLFlow experiment tracking and comparison are all provided while allowing you to track the model and data versioning for reproducibility, audits and governance.
Installation
Requirements
The following is the minimum configuration required to deploy DKube on a Rancher cluster
- The minimal configuration for each of the worker nodes is as follows:
- 16 cores
- 64 GB RAM
- 300 GB storage for Root Volume
- The worker nodes could be brought up with any of the following OS distributions
- Ubuntu 20.04
- CentOS / RHEL 7.9
- Amazon Linux 2 for installations on AWS
- Storage
- The recommended storage option for DKube meta-data and user ML resources is an external NFS server with a min of 1TB storage available.
- For evaluation purposes, one of the worker nodes can be configured as the storage option. In this case the recommended size of storage on the worker node is 1 TB and a minimum size of 400 GB.
- Dkube requires a Kubernetes version of 1.20.
- Dkube images registry details are required for installation. Please send a mail to support@dkube.io for the details.
- The following sections in the installation guide needs to be followed to prepare Rancher cluster for Dkube installation.
- Getting the Dkube Files
- Setting up the Rancher Cluster
- Preparing the Rancher Cluster.
- Node Setup. This is optional for a non-GPU cluster.
For more information on installation, refer to the Dkube Installation Guide.