Describe how you would implement a distributed training pipeline using PyTorch's DistributedDataParallel (DDP).