Pytorch Error, RuntimeError: expected scalar type Long but found Double

user_8613875 0
Viewed 3027

I have run into the following error while training a BERT classifier. The 

type(b_input_mask) = type(b_labels) = torch.Tensor      

type(b_labels[i]) = tensor(1., dtype=torch.float64)

type(b_input_masks[i]) = class'torch.Tensor'

What could be the possible data type error here since I have not typecasted any variable to either long or double?

Thanks in advance! Error Stack Trace

pytorch python types
2 Answers

In a classification task, the data type for input labels should be Long but you assigned them as float64

type(b_labels[i]) = tensor(1., dtype=torch.float64)

=>

type(b_labels[i]) = tensor(1., dtype=torch.long)
user_62982020

You can use torch.Tensor.long to convert tensor to expected long type.

# Here, you can pass parameter like this in your call
..., labels = b_labels.long())
user_60756990
Related