The remaining region workloads move to ghcr.io/brf-tech/runr,
dedicated-worker tolerations, organization quota overcommit, hf-serve
1.1.1 and the TensorFlow -gpu runtime on the GPU.