Convert ResNet to Sequential?

Part 1 (2019)
1 
def custom_resnet(pretrained=False):
    model = torchvision.models.resnet50(pretrained=pretrained)
    model = torch.nn.DataParallel(model).cuda()    
    checkpoint = model_zoo.load_url(x)       
    model.load_state_dict(checkpoint["state_dict"])    
    
    print(type(model)) # <class 'torch.nn.parallel.data_parallel.DataParallel'>
    
    all_layers = list(model.children())    

    print(all_layers) # [ResNet((conv1): ... (fc): Linear(in_features=2048, out_features=1000, bias=True))]
    
    return nn.Sequential(*all_layers[0], *all_layers[1:])

Hey guys! The above is my custom pretrained resnet and I am passing to the cnn_leaner as follows.

  learn = cnn_learner(src, 
                        base_arch=custom_resnet,
                        pretrained=True,
                        metrics=error_rate)

I am getting the error message of type object argument after * must be an iterable, not ResNet. I guess I will have to convert the ResNet to Sequential?

Any help is appreciated. Thank you.

Edit: if i only did return nn.Sequential(model), the error thrown is

cut = next(i for i,o in reversed(ll) if has_pool_type(o))

StopIteration:
1 Like