DEVOPSdigest

Red Hat announced advances in Red Hat OpenShift AI, an open hybrid artificial intelligence (AI) and machine learning (ML) platform built on Red Hat OpenShift that enables enterprises to create and deliver AI-enabled applications at scale across hybrid clouds.

These updates highlight Red Hat’s vision for AI, bringing Red Hat’s commitment to customer choice to the world of intelligent workloads, from the underlying hardware to the services and tools, such as Jupyter and PyTorch, used to build on the platform. This provides faster innovation, increased productivity and the capacity to layer AI into daily business operations through a more flexible, scalable and adaptable open source platform that enables both predictive and generative models, with or without the use of cloud environments.

Red Hat’s AI strategy enables flexibility across the hybrid cloud, provides the ability to enhance pre-trained or curated foundation models with their customer data and the freedom to enable a variety of hardware and software accelerators. Red Hat OpenShift AI’s new and enhanced features deliver on these needs through access to the latest AI/ML innovations and support from an expansive AI-centric partner ecosystem.

The latest version of the platform, Red Hat OpenShift AI 2.9, delivers:

- Model serving at the edge extends the deployment of AI models to remote locations using single-node OpenShift. It provides inferencing capabilities in resource-constrained environments with intermittent or air-gapped network access. This technology preview feature provides organizations with a scalable, consistent operational experience from core to cloud to edge and includes out-of-the-box observability.

- Enhanced model serving with the ability to use multiple model servers to support both predictive and GenAI, including support for KServe, a Kubernetes custom resource definition that orchestrates serving for all types of models, vLLM and text generation inference server (TGIS), serving engines for LLMs and Caikit-nlp-tgis runtime, which handles natural language processing (NLP) models and tasks. Enhanced model serving allows users to run predictive and GenAI on a single platform for multiple use cases, reducing costs and simplifying operations. This enables out-of-the-box model serving for LLMs and simplifies the surrounding user workflow.

- Distributed workloads with Ray, using CodeFlare and KubeRay, which uses multiple cluster nodes for faster, more efficient data processing and model training. Ray is a framework for accelerating AI workloads, and KubeRay helps manage these workloads on Kubernetes. CodeFlare is central to Red Hat OpenShift AI’s distributed workload capabilities, providing a user-friendly framework that helps simplify task orchestration and monitoring. The central queuing and management capabilities enable optimal node utilization, and enable the allocation of resources, such as GPUs, to the right users and workloads.

- Improved model development through project workspaces and additional workbench images that provide data scientists the flexibility to use IDEs and toolkits, including VS Code and RStudio, currently available as a technology preview, and enhanced CUDA, for a variety of use cases and model types.

- Model monitoring visualizations for performance and operational metrics, improving observability into how AI models are performing.

- New accelerator profiles enable administrators to configure different types of hardware accelerators available for model development and model-serving workflows. This provides simple, self-service user access to the appropriate accelerator type for a specific workload.