Citing Forrester research that found 76 per cent of data scientists and IT practitioners expect ML usage to increase over the next couple of years, D2iQ has emitted an enterprise-ready distribution of Kubeflow in the form of KUDO.
Kubeflow itself is a cloud-native platform (thus ticking the Kubernetes buzzword bingo box) for ML based on Google’s internal ML pipelines. The theory goes that anywhere you are running Kubernetes, you should be able to run Kubeflow for your ML requirements.
There are multiple tools, apps and scaffolding needed, not least Kubeflow’s own operators, as well as the likes of Jupyter and PyTorch to make the magic happen, from the initial experimental phase and early model training to the online prediction and ongoing training of the production phase.
Snuggling under D2iQ’s Ksphere umbrella, where end-to-end support for Kubernetes can be found, KUDO puts the toys together for ML engineers and data scientists, replete with pre-installed Jupyter Notebooks and frameworks for distributed training.
The thinking is that as well as upping productivity, a single platform will also make life easier when deploying and training ML models at scale as well as reduce the risk associated with something a bit more manual.
Headquartered in San Francisco, D2iQ is all about “Day 2 operations” – that wonderful time when a system is ticking over, having been through the pain of getting started in the first place. It rarely refers to an actual second day (although having seen many agile methodologies overpromise and underdeliver over the years, we’re pretty sure that somewhere along the line some consultant has assured a gullible manager that two days are all that are needed).
Getting to that glorious Day 2 point can be tricky for ML, and stirring Kubernetes into the mix can crank the complexity bar up a notch or two. Hence the simplifications of Kubeflow packaged up in the enterprise-ready KUDO.
D2iQ added cloud native CI/CD platform Dispatch to its Kubernetes distribution, Konvoy, back in March.