AttributeError: module 'torch._C' has no attribute '_cuda_getDevice

[ronyuzhang]

i train my own data by using your checkpoint of checkpoint_unets , but have a error "AttributeError: module 'torch._C' has no attribute '_cuda_getDevice".
i train unet on cpu enviroment.

[pgtgrly]

I have made some changes, Can you pull the repository and test again? Please update me if it works or not.

[pgtgrly]

Closing as the issue is resolved

[PeterXiaoGuo]

Hi, @pgtgrly

I face the same problem when I re-produce another code when I loading the .pt files.

I have no Nvidia GPU and run on a cpu only computer

Could you please share how you fix the "module 'torch._C' has no attribute '_cuda_getDevice" problem? Thank you very much!

Best,
Xiao

[pgtgrly]

Hey @PeterXiaoGuo
To load the GPU trained weights to CPU you will have to map them. This can be done by simply loading the model using:

model=torch.load(trained_model.pt,map_location={'cuda:0': 'cpu'})

You can find the same in my code here

Hope this helps.
Regards
PG

[PeterXiaoGuo]

Hey, @pgtgrly
Thank you very much!!!
I made it! Thank you !

Best,
Xiao

[pgtgrly]

@PeterXiaoGuo,
I am glad that I could help :)
Good Luck
PG

[ZangChaoyu]

我有GPU ，怎样解决呢请问