The unprecedented promise of machine learning (ML) is still unrealized, because data scientists are spending most of their time on non-data-science work. The common practice is that ML development through deployment relies on ad hoc tools, plug-ins, scripts, and a myriad of siloed tools that are impeding organizations, large and small, from streamlining ML development.
NetApp and cnvrg.io have partnered to deliver an AI/ML data science pipeline solution that is streamlined and drives productivity and efficiency. The solution incorporates industry-leading Kubernetes managed clusters (for example, Red Hat OpenShift), cached datasets for extreme performance, and the one-click attachments of models to datasets with NVIDIA NGC integration. NetApp® ONTAP® AI provides high-performance compute and storage for any scale of operation, and cnvrg.io software streamlines data science workflows, improving resource utilization.
With NetApp and cnvrg.io, you can cache datasets (and/or their versions) and make sure that they’re located in the ONTAP node attached to the GPU cluster or CPU cluster that is exercising the training. Once the datasets are cached, they can be used multiple times by different team members. With caching, datasets are ready to be used in seconds rather than hours, and cached datasets can be authorized and used by multiple teams in the same compute cluster connected to the NetApp cached data.
NetApp and cnvrg.io have written a detailed technical paper, Hybrid-cloud AI Operating System with Data Caching, which presents an innovative solution that enables IT professionals and data engineers to create a truly hybrid-cloud AI platform with a topology-aware data hub. Data scientists can instantly and automatically create a cache of their datasets in proximity to the compute, wherever the compute is located. As a result, high-performance model training can be easily accomplished and multiple AI practitioners can collaborate with immediate access to the cached datasets and versions, and with the ability to create a dataset-version hub.