OSError: [Errno 22] Invalid argument:

[Feez]

examples/fast_neural_style/neural_style/neural_style.py

Line 117 in 36441a8

torch.save(transformer.state_dict(), save_model_path)

Spent 2 hours training a model for it to fail the save with error prefixed with OSError: [Errno 22] Invalid argument :(

For now I'm just adding fail-safe but I am sure it can be written cleaner

custom_save_path = save_model_path
while True:
    try:
        torch.save(transformer.state_dict(), custom_save_path)
        break
    except:
        print("Please enter new save path:")
        custom_save_path = input() 

print("\nDone, trained model saved at", custom_save_path)

[msaroufim]

@Feez would you like to make a PR for this? Could you also show us the full error you're getting?

[Feez]

I'm sorry I don't have this environment set up anymore but I believe the error was a result of just inserting a save path that is syntactically valid but didn't actually exist. Such as a save path in a directory that doesn't exist.

Regardless I think there probably should be more fail safes around the actual saving of the file to prevent loss of progress in training a model, as I imagine this could happen with many other OS related issues that the user might want to manually resolve and attempt to re-save (e.g. out of disk space on save path, drive unplugged during save, network path disconnected, etc)

[Feez]

Created a quick PR to address the general problem stated above. Hopefully this suffices.