pytorch - pow not implemented for Half

Question

I have this code part for training a Bert model:

from fastai.callbacks import *

learner = Learner(
    databunch, bert_model,
    loss_func=loss_func,
)

learner.lr_find() 

and facing this runtime error

/usr/local/lib/python3.6/dist-packages/pytorch_pretrained_bert/modeling.py in forward(self, x)
    240         def forward(self, x):
    241             u = x.mean(-1, keepdim=True)
--> 242             s = pow((x - u),2).mean(-1, keepdim=True)
    243             x = (x - u) / torch.sqrt(s + self.variance_epsilon)
    244             return self.weight * x + self.bias

RuntimeError: "pow" not implemented for 'Half'

How can I solve this issue? Any help would be great!

Thanks

Answer 1

等待大神答复