how to apply DeepSpeedExamples with mpirun ?

Hi i have been using mpi-operator, and found an interesting project according to https://github.com/dogukanutuna/mpi-operator/tree/master/examples/v2beta1/deepspeed which can apply cifar10_deepspeed.py with mpirun of mpi-operator on k8s cluster, and distributed training can be executed successfully for me.
So i wonder, can all of these script in DeepSpeedExamples/training be applyed by this mpi-operator-way with simple command mpirun ?
In my case, at least, gan_deepspeed_train.py can not be applied by such way with mpirun.
There were somethings about distributed training i found, however not sure whether they are correct or not:
When submitting command "deepspeed script.py --deepspeed". args.local_rank would be assigned implicitly according to my env before deepspeed.init_distributed(), I think it's critical for distributed training.
But when submitting "mpirun python script.py --deepspeed", args.local_rank can only be passed externally with manual parameter,
which goes against the convenience of mpi-operator.
any hint would be helpful to me. Thank you in advance~

Start with the DeepSpeedExamples/training scripts mentioned in the issue, especially gan_deepspeed_train.py, and compare their invocation with cifar10_deepspeed.py under mpi-operator. Determine whether the examples support mpirun and how local_rank is supplied; the issue does not define a specific code change or acceptance criterion.