Pytorch 1.1 load model from pytorch 0.4

vision
1

Hey guys, Recently I met a problem.
The case is I want to load a model with pytorch 1.1 , the model is trained by pytorch 0.4. I cannot make it to load the model correctly.
Can anybody meet this problem before, how to solve this problem?
Thx for ur kindly help~

2

Do you get a specific error message?
0.4.0 is quite old by now, so loading the model might not work out of the box, but we would need some kind of error in order to help debugging.

3

Sry for missing the error message.
Here is the meaasge goes: “‘LogSoftmax’ object has no attribute ‘_state_dict_hooks’”

4

Are you trying to load the state_dict or the model via torch.load('model.pyt') when getting this error?

5

Yes. I load model weights from ‘***.pth’ , then when I replace ‘module.’ with ‘’ , this error occured .

6

The replacement of .module points to loading a state_dict from a nn.DataParallel model and should most likely be unrelated to the error.
Do you see any values in your loaded state_dict for the LogSoftmax layer?