Saving the right model for testing

Hi all,

I don’t know if this pytorch-related forum is the right to ask this kind of question I hope is ok.

I have seen several implementations in deep learning that save the model with the minimum validation loss and use this model for testing but I have noticed that my training loss is even smaller, is right to use the model with the minimum training loss? maybe my question is silly bus I am newbie in this area.

Also, using the following command I save where? the model, in which directory

torch.save(model.state_dict(), './state_dict.pt')

Usually you would save the model with the lowest validation loss, since the lowest training loss could yield to overfitting.

The path (and directory) is given as the second argument. In your example, you are using the current working directory via ./