pytorch 0.4 will make error: RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation about bdrar

[Jee-King]

but infer.py must run in pytorch-0.4 ....

[zijundeng]

Hi,
You could cancel the inplace behavior of dropout in the model.py.
Just modify the code

for m in self.modules():
    if isinstance(m, nn.ReLU) or isinstance(m, nn.Dropout):
        m.inplace = True

to be

for m in self.modules():
    if isinstance(m, nn.ReLU):
        m.inplace = True

[Jee-King]

Thank you very much!