RuntimeError: expected scalar type Double but found Float while tensor is already double

1

I have come across a very usual error RuntimeError: expected scalar type Double but found Float. But the tensor is already double. So probably it comes from some different place which I don’t see…

It comes from here

h = self.batch_norms[layer](h)

Again h is float64 (or double) and layer is int.
By the way, self.batch_norms is simple toch.nn.ModuleList, i.e. self.batch_norms = torch.nn.ModuleList() declared in constructor.

Here is a full traceback

RuntimeError                              Traceback (most recent call last)
C:\Users\ROSTYS~1\AppData\Local\Temp/ipykernel_6476/2774919388.py in <module>
      1 for data in loader:
----> 2     out = model(data.x, data.edge_index, deg)

~\anaconda3\envs\resal\lib\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 = [], []

C:\Users\ROSTYS~1\AppData\Local\Temp/ipykernel_6476/730682502.py in forward(self, x, edge_index, deg)
     58             h = self.convs[layer](h=h, edge_index=edge_index, deg = deg)
     59             h = self.activation(h)
---> 60             h = self.batch_norms[layer](h)
     61 
     62         if self.task == "node":

~\anaconda3\envs\resal\lib\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\envs\resal\lib\site-packages\torch\nn\modules\batchnorm.py in forward(self, input)
    165         used for normalization (i.e. in eval mode when buffers are not None).
    166         """
--> 167         return F.batch_norm(
    168             input,
    169             # If buffers are not to be tracked, ensure that they won't be updated

~\anaconda3\envs\resal\lib\site-packages\torch\nn\functional.py in batch_norm(input, running_mean, running_var, weight, bias, training, momentum, eps)
   2279         _verify_batch_size(input.size())
   2280 
-> 2281     return torch.batch_norm(
   2282         input, weight, bias, running_mean, running_var, training, momentum, eps, torch.backends.cudnn.enabled
   2283     )

RuntimeError: expected scalar type Double but found Float

Does somebody know, where is the problem?

Thanks!

2

Solution was to make sure that the model parameters are also double, i.e. model.double() helped

1 Like