RuntimeError with convert_model - "found one of them on device: cpu" about synchronized-batchnorm-pytorch

[FreemanG]

Or maybe what really matters is to attach .to(device) in the end...

[FreemanG]

I did some search in the issues, then changed the codes to:

device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
model = net()  # my model
model = convert_model(model)

if args.use_gpu:
    # model = torch.nn.DataParallel(model, device_ids=gpus).to(device)
    model  = DataParalleWithCallBack(model, gpus).to(device)

The codes start to run without error, although not sure about the performance. It seems that I have to use your DataParallelWithCallback instead of pytorch nn.DataParallel, like the issue 19