Optimizing AI workloads in Kubernetes: Pruning for efficiency and scale