Stable Diffusion in the Classroom: Deploying interactive GPU-enabled ML workloads with Open OnDemand and Kubernetes

This paper explores the deployment of Stable Diffusion instances in educational settings, particularly focusing on addressing resource allocation challenges in the context of interactive GPU-enabled machine learning (ML) workloads. Traditional approaches to providing GPU resources in High-Performance Computing (HPC) environments often involve assigning a full GPU to each user, which can be cost-prohibitive and inefficient for students in a classroom. To overcome this limitation, we deploy with a novel solution leveraging Kubernetes (k8s) and GPU sharing through Open OnDemand, a platform for accessing HPC resources via web interfaces. Our innovative approach demonstrates the feasibility and effectiveness of employing Kubernetes and Open OnDemand to deploy interactive GPU-enabled ML workloads in classroom environments.