The CPU runtime was pinned and configured for MKL-DNN in a non-root container, but installation and inference were not exercised because no container runtime or baked model image was available.
Partly done
Running OCR inference on CPU workers
Got in the wayInstallationConfiguration