b5f41f828b0ec9b67fa60aceb0778073d1b368b2,fairseq/distributed_utils.py,,distributed_init,#Any#,71

Before Change

            socket.gethostname(), args.distributed_rank), flush=True)

        // perform a dummy all-reduce to initialize the NCCL communicator
        dist.all_reduce(torch.zeros(1).cuda())

        suppress_output(is_master(args))

    args.distributed_rank = torch.distributed.get_rank()

After Change

            socket.gethostname(), args.distributed_rank), flush=True)

        // perform a dummy all-reduce to initialize the NCCL communicator
        if torch.cuda.is_available():
            dist.all_reduce(torch.zeros(1).cuda())
        else:
            dist.all_reduce(torch.zeros(1))

        suppress_output(is_master(args))

    args.distributed_rank = torch.distributed.get_rank()
    return args.distributed_rank

Instances