Different uses of torch.save

How can we benefit afterwards from saving both model and [args.arch, model.state_dict()] ?

# Final model
    torch.save(model, MODEL_DIR / f"final_{args.fold}.pth")
    # Failover (args + state dict)
        [args.arch, model.state_dict()],
        MODEL_DIR / f"failover_{args.arch}_{args.fold}.pth"

I would recommend to save the state_dict instead of the model directly as explained here.
I’m not sure, what args.arch is defined, but I guess it would be a string to a pretrained torchvision model?

Yes it is a string of the name of the architecture, like “densenet121” for example.