well just run all inference on the cpu, single threaded /s