How to load a parameter file saved as zip

1

Hello,

I have two version of pytorch. The old one installed on my PC computer, and the new one on Linux server. I didn’t realize the change on the function torch.save, so the parameters trained on server were saved as zip files. When I downloaded them and tried to evaluate on my PC, some running error occurred. Is the torch.save and net.load_state_dict in two versions incompatible? What I should do to use my training results on server?

Thank you !

1 Like
2

Hi,
Had the same issue. I don’t know what the correct fix would be, but as a temporary fix you could load the saved .zip files on your server and then save them again with torch.save(obj, f, _use_new_zipfile_serialization=False).
This will give you the old format.
You can read more about this in the notes here

1 Like
3

Thank you very much for the solution :blush: