RuntimeError: Input type (torch.cuda.FloatTensor) and weight type (torch.cuda.DoubleTensor) should be the same

Hi, I’m trying to train a pre-trained model, with my GPU. Everything worked well but I wanted to code a class in order to have a cleaner code. On my class I do the same operations that I was doing without my class, but I have an issue that I don’t understand, with the line model(images, targets) :

---------------------------------------------------------------------------
RuntimeError                              Traceback (most recent call last)
<ipython-input-10-6af86cd2bf74> in <module>
----> 1 class_model.train(num_epoch = 1)

<ipython-input-6-700e8ee37055> in train(self, num_epoch, gpu)
    133 
    134                 # Train for one epoch, printing every 10 iterations
--> 135                 train_his_, list_losses, list_losses_dict = train_one_epoch(model, optimizer, self.data_loader, device, epoch, print_freq=10)
    136                 list_of_list_losses.append(list_losses)
    137                 # Compute losses over the validation set

<ipython-input-2-11a7da6d9e67> in train_one_epoch(model, optimizer, data_loader, device, epoch, print_freq)
    508 
    509         # Feed the training samples to the model and compute the losses
--> 510         loss_dict = model(images, targets)
    511         losses = sum(loss for loss in loss_dict.values())
    512 

~/anaconda3/lib/python3.8/site-packages/torch/nn/modules/module.py in _call_impl(self, *input, **kwargs)
   1049         if not (self._backward_hooks or self._forward_hooks or self._forward_pre_hooks or _global_backward_hooks
   1050                 or _global_forward_hooks or _global_forward_pre_hooks):
-> 1051             return forward_call(*input, **kwargs)
   1052         # Do not call functions when jit is used
   1053         full_backward_hooks, non_full_backward_hooks = [], []

~/anaconda3/lib/python3.8/site-packages/torchvision/models/detection/generalized_rcnn.py in forward(self, images, targets)
     91                                      .format(degen_bb, target_idx))
     92 
---> 93         features = self.backbone(images.tensors)
     94         if isinstance(features, torch.Tensor):
     95             features = OrderedDict([('0', features)])

~/anaconda3/lib/python3.8/site-packages/torch/nn/modules/module.py in _call_impl(self, *input, **kwargs)
   1049         if not (self._backward_hooks or self._forward_hooks or self._forward_pre_hooks or _global_backward_hooks
   1050                 or _global_forward_hooks or _global_forward_pre_hooks):
-> 1051             return forward_call(*input, **kwargs)
   1052         # Do not call functions when jit is used
   1053         full_backward_hooks, non_full_backward_hooks = [], []

~/anaconda3/lib/python3.8/site-packages/torchvision/models/detection/backbone_utils.py in forward(self, x)
     42 
     43     def forward(self, x):
---> 44         x = self.body(x)
     45         x = self.fpn(x)
     46         return x

~/anaconda3/lib/python3.8/site-packages/torch/nn/modules/module.py in _call_impl(self, *input, **kwargs)
   1049         if not (self._backward_hooks or self._forward_hooks or self._forward_pre_hooks or _global_backward_hooks
   1050                 or _global_forward_hooks or _global_forward_pre_hooks):
-> 1051             return forward_call(*input, **kwargs)
   1052         # Do not call functions when jit is used
   1053         full_backward_hooks, non_full_backward_hooks = [], []

~/anaconda3/lib/python3.8/site-packages/torchvision/models/_utils.py in forward(self, x)
     60         out = OrderedDict()
     61         for name, module in self.items():
---> 62             x = module(x)
     63             if name in self.return_layers:
     64                 out_name = self.return_layers[name]

~/anaconda3/lib/python3.8/site-packages/torch/nn/modules/module.py in _call_impl(self, *input, **kwargs)
   1049         if not (self._backward_hooks or self._forward_hooks or self._forward_pre_hooks or _global_backward_hooks
   1050                 or _global_forward_hooks or _global_forward_pre_hooks):
-> 1051             return forward_call(*input, **kwargs)
   1052         # Do not call functions when jit is used
   1053         full_backward_hooks, non_full_backward_hooks = [], []

~/anaconda3/lib/python3.8/site-packages/torch/nn/modules/conv.py in forward(self, input)
    441 
    442     def forward(self, input: Tensor) -> Tensor:
--> 443         return self._conv_forward(input, self.weight, self.bias)
    444 
    445 class Conv3d(_ConvNd):

~/anaconda3/lib/python3.8/site-packages/torch/nn/modules/conv.py in _conv_forward(self, input, weight, bias)
    437                             weight, bias, self.stride,
    438                             _pair(0), self.dilation, self.groups)
--> 439         return F.conv2d(input, weight, bias, self.stride,
    440                         self.padding, self.dilation, self.groups)
    441 

RuntimeError: Input type (torch.cuda.FloatTensor) and weight type (torch.cuda.DoubleTensor) should be the same

My device is already cuda:0.
The data are called on this function basically from the engine librairie. I tried to cast into a torch.float but it didn’t work


def train_one_epoch(model, optimizer, data_loader, device, epoch, print_freq):


    model.train()
    metric_logger = utilss.MetricLogger(delimiter="  ")
    metric_logger.add_meter('lr', utilss.SmoothedValue(window_size=1, fmt='{value:.6f}'))
    header = 'Epoch: [{}]'.format(epoch)
    list_losses = []
    list_losses_dict = []
    for i, values in tqdm(enumerate(metric_logger.log_every(data_loader, print_freq, header))):
        images, targets = values
        print(type(images))
        images = list(image.to(device, dtype=torch.float) for image in images)
        print(type(images))
        targets = [{k: v.to(device, dtype = torch.float) for k, v in t.items()} for t in targets]
        #images = [image.cuda() for image in images]

        # Feed the training samples to the model and compute the losses
        loss_dict = model(images, targets)
        losses = sum(loss for loss in loss_dict.values())

        # reduce losses over all GPUs for logging purposes
        loss_dict_reduced = utilss.reduce_dict(loss_dict)
        losses_reduced = sum(loss for loss in loss_dict_reduced.values())
        loss_value = losses_reduced.item()
        print("Loss is {}, stopping training".format(loss_value))
        if not math.isfinite(loss_value):
            print("Loss is {}, stopping training".format(loss_value))
            print(loss_dict_reduced)
            sys.exit(1)
        list_losses.append(loss_value)

        # Pytorch function to initialize optimizer
        optimizer.zero_grad()
        # Compute gradients or the backpropagation
        losses.backward()
        # Update current gradient
        optimizer.step()

        metric_logger.update(loss=losses_reduced, **loss_dict_reduced)
        metric_logger.update(lr=optimizer.param_groups[0]["lr"])

        # Record losses to plot learning curves
        if i == 0: 
            history = {key: val.cpu().detach() for key, val in loss_dict_reduced.items()}
            history['loss'] = losses_reduced.cpu().detach()
        else:
            for key, val in loss_dict_reduced.items():history[key] += val.cpu().detach()
            history['loss'] += losses_reduced.cpu().detach()
        
         
        torch.save(model.state_dict(), FILE_model_dict_gpu)
        list_losses_dict.append(loss_dict_reduced)
        save_obj(history, "history_train_fixed_frame_lab2_and_lab7_5epoch_07-07")
        plt.plot([x for x in range(len(list_losses))],list_losses)
        plt.show()   
    return history, list_losses, list_losses_dict

Thanks for your help !

1 Like

Based on the error message (some) model parameters are DoubleTensors, while the input is a FloatTensor, so you would either have to leave the model parameters in the default dtype (or call model.float() explicitly) or alternatively transform the inputs to DoubleTensors via input = input.double().

1 Like

Thank you ! I converted every input into float and it worked

1 Like