How to start multigpu training from within a function?

[Revist]

Looking at train_net.py we start multigpu training as:

if __name__ == "__main__":    
  args = default_argument_parser().parse_args()    
print("Command Line Args:", args)    
launch(main,  args.num_gpus,  num_machines=args.num_machines, machine_rank=args.machine_rank,        dist_url=args.dist_url, args=(args,),    )

What if I need to start multigpu training from within a function where I have some variables holding my dataset (like the usual detectron2 format dataset of list of dictionaries)? If I did not need to place launch within "if name == "main": " and could easily pass the dataset that would be easy. Is there any easy way to circumvent this problem?