Data parallelism error for pretrained model

1

Hi,

I am trying to run my model on multiple GPUs for data parallelism but receiving this error:

AttributeError: 'DataParallel' object has no attribute 'fc'

I have defined the following pretrained model :

def resnet50(num_classes, device):
    model = models.resnet50(pretrained=True)
    model = torch.nn.DataParallel(model)
    for p in model.parameters():
        p.requires_grad = False
    num_ftrs = model.fc.in_features
    model.fc = nn.Linear(num_ftrs, num_classes)

    model = model.to(device)
    return model

It’s unclear to me where I can add “module”

Cheers

2

If you are trying to access the fc layer in the resnet50 wrapped by the DataParallel model, you can use model.module.fc, as DataParallel stores the provided model as self.module:

3

Great, thanks. I added .module to everything before .fc including the optimizer.