Gpu trained parameters get different when loading on cpu

[newman_chiang]

Hi,

I’m using torch 1.8.1+cu101.
I train model on gpu and save it using torch.save(…) and load it back on cpu using torch.load(…, map_location=‘cpu’).
But the prediction result on cpu is totally different from that on gpu.
I then check the model parameters loaded and the parameters are different on cpu and on gpu.
Why is it?

I’ve already seen

On a cpu device, how to load checkpoint saved on gpu device

I trained my network on a gpu device and saved checkpoint by torch.save Loading this checkpoint on my cpu device gives an error: raise AssertionError("Torch not compiled with CUDA enabled") AssertionError: Torch not compiled with CUDA enabled```

but it still doesn’t work.

Thanks

[ptrblck]

Could you post the model definition as well as a minimal code snippet to reproduce the issue of the non-matching parameters, please?