lightning:
  trainer:
    distributed_backend: ddp
    accumulate_grad_batches: 4
    gpus: 0,1