Removed assert in case of external ddp, exported initialize_megatron()

from .global_vars import get_tensorboard_writer

from .global_vars import get_adlr_autoresume

from .global_vars import get_timers

from .initialize  import initialize_megatron

def print_rank_0(message):

    """If distributed is initialized print only on rank 0."""

                  'skipping initialization ...', flush=True)

        args.rank = torch.distributed.get_rank()

        args.world_size = torch.distributed.get_world_size()

        if device_count > 0:

            device = torch.cuda.current_device()

            local_rank = args.rank % device_count

            assert local_rank == device, \

                'expected local-rank to be the same as rank % device-count.'

    else: