CPU count during training

yes, num_processes is used only with ddp_cpu as distributed backend, so in 1.2 we will set the num always to max unless you define any other number; for now, you can use num_processes=os.cpu_count()