Open-source software giant Red Hat Inc. is upping its game in artificial intelligence development with a host of updates to the Red Hat OpenShift AI platform, announced at KubeCon + CloudNativeCon North America 2024 today in Salt Lake City, Utah.

Red Hat OpenShift AI is a scalable AI and machine learning development platform that enables enterprises to create and deliver AI-powered applications at scale across hybrid cloud environments. In other words, they can use it to create AI models and applications that tap into their data, wherever it resides, in the cloud or on-premises.

With today’s update, the IBM Corp.-owned company is adding multiple new features that will allow organizations to turbocharge the AI capabilities within their existing, mission-critical applications.

Red Hat OpenShift AI 2.15 will become available later this month, and the headline feature is a new model registry, available as a technology preview, that provides a centralized location for users to view and manage registered AI models. Employing predictive and generative AI capabilities, the registry provides users with a structured way to organize, share, version, deploy and track their models, as well as the underlying metadata and model artifacts they rely on.

Meanwhile, the new data drift detection monitor works by detecting changes in the input data distribution of machine learning models that are up and running in production, so it can alert users when the live data used for model inference significantly deviates from the information it was originally trained on. This is an important capability that helps to maintain the reliability of AI models, Red Hat said, ensuring they remain aligned with their original training to maintain the accuracy of their predictions and responses.

The new bias detection tools do much the same, however they’re more focused on AI model fairness. The idea is that they can prevent cases of AI bias. This is a common problem for AI models. For instance, an AI that’s tasked with filtering job applicants might exhibit bias against candidates of a specific ethnicity or educational background, and the new feature is designed to prevent this.

AI developers are also getting more tools to enhance model fine-tuning. Red Hat OpenShift AI 2.15 adds support for LoRA, or low-rank adapters, which make it possible to fine-tune LLMs such as Meta Platforms Inc.’s Llama 3, in order to limit resource consumption and make them more affordable to run at large scale.

Other enhancements announced today include the integration of Nvidia Corp.’s NIM microservices, which provide access to a wide selection of easy-to-deploy large language models for generative AI applications, and support for graphics processing units made by Advanced Micro Devices Inc., giving customers more options in terms of AI processing hardware. There are also new model serving features for developers to look forward to, such as vLLM serving runtimes for KServe, and expanded AI training and experimentation options such as the hyperparameter tuning tool Ray Tune.

Joe Fernandes, vice president and general manager of Red Hat’s AI business unit, said organizations need a reliable, scalable and flexible platform for AI that can run wherever their data lives. “We’re offering significant improvements in scalability, performance and operational efficiency… making it possible for IT organizations to gain the benefits of a powerful AI platform while maintaining the ability to build, deploy and run on whatever environment their unique business needs dictate.”

Next-gen Red Hat OpenShift enters preview

Red Hat’s focus on AI is not just about helping companies to develop and deploy intelligent applications. In addition, it’s also using AI to enhance the capabilities of its own platforms, such as the classic Red Hat OpenShift, a popular Kubernetes platform that’s used by developers to build and deploy distributed and scalable applications across any cloud or on-premises environment.

The Red Hat OpenShift Lightspeed platform, available now as a technology preview, represents the culmination of those AI integration efforts, introducing a new generative AI-powered assistant that lets users ask it questions in their natural language.

Red Hat said the OpenShift virtual assistant is designed to assist with tasks like troubleshooting applications and investigating cluster resources, with the goal to simplify life for users and make them more productive.

While Lightspeed is still only in preview, the latest Red Hat OpenShift 4.17 release is generally available now, although the new features are a little less exciting. The main focus of the update has been on improving virtualized workload management with enhancements to Red Hat OpenShift Virtualization. In addition, the company announced advanced Kubernetes management capabilities, making it easier to manage virtual machines across multiple clusters.

AI application templates

Coming back to AI, Red Hat announced a bunch of new features in Red Hat Developer Hub, which is an enterprise-grade developer portal based on the open-source Backstage project.

With this, Red Hat is introducing new AI-focused software templates for developers to follow a more standardized approach to creating and deploying AI-enhanced services and components. The new templates cover common applications such as audio-to-text, chatbots, code generation, object detection and retrieval-augmented generation. By using the templates, developers get what amounts to a prebuilt application that can then be customized to their liking, saving them time on development.

Finally, the company announced Red Hat Device Edge 4.17, with updates aimed at easing the deployment of AI applications that live at the edge, on low-powered devices such as sensors, drones and robots. Red Hat Device Edge is a platform for resource-constrained edge environments, where small form factor devices and compute resources require lower-latency operations.

Image: SiliconANGLE/Freepik AI