Red Hat Extends Managed OpenShift Service for Building AI Models
Red Hat extended the capabilities of its managed service for building artificial intelligence (AI) models on top of its Kubernetes distribution to foster greater collaboration among data scientists and DevOps teams.
In addition, the latest edition of Red Hat OpenShift Data Science service includes support for NVIDIA data center graphical processor units (GPUs) that can run a certified instance of the NVIDIA AI Enterprise software suite.
Red Hat is also now giving organizations the option to use unique images and custom Jupyter Notebooks versus the various open source tools and libraries curated by Red Hat. At the same time, Red Hat is making available a portal through which developers can learn to use these tools.
Finally, Red Hat OpenShift Data Science is now available on the Amazon Web Services (AWS) marketplace. It is now integrated with Pachyderm, a data versioning tool that makes it possible to use containers to build machine-learning pipelines with guaranteed data lineages.
Red Hat OpenShift Data Science is scheduled to be generally available in the next few weeks after being initially launched late last year.
Will McGrath, product marketing manager, Cloud Services, Red Hat, said one of the challenges that Red Hat is enabling organizations to better address is the current divide between data science and DevOps teams. In many cases, he says, data science teams are using a separate set of machine learning operations (MLOps) platforms that should ideally be deployed on the same Kubernetes platforms being used to build cloud-native applications.
That creates an opportunity to bring those diverse cultures together in a way that allows distinct sets of processes to be more easily melded, McGrath adds.
The bulk of AI applications are being built using containers. Otherwise, many of these applications would be too unwieldy to construct given the volume of data required to train an ML model. As AI continues to evolve, it’s still unclear whether MLOps and DevOps practices will converge. Data science teams today typically have defined their own workflow processes using a wide range of graphical tools. However, as it becomes obvious that almost every application will be infused with machine learning and deep learning algorithms, there is a growing need to bridge the current divide between DevOps and data science teams.
DevOps teams should assume that there will not only be many more AI models, but that those models will need to be continuously updated. This is because each AI model is built based on a set of assumptions; as more data becomes available and those assumptions change, the AI models are subject to drift that results in less accuracy over time. Organizations may determine that an entire AI model needs to be replaced because the business conditions on which assumptions were originally made are no longer valid.
One way or another, updating and tuning AI models is likely to become another continuous process managed via a DevOps workflow. The only thing left to determine is how much friction will be encountered in the process of achieving that goal.
