Is it possible to save best model only ONE time after training process?

[GillianGrayson]

Hi!

Is it possible to save the best model only ONE time after training process (EarlyStopping callback)?

If I understand correctly, ModelCheckpoint allows you to save the best model only in iterative mode.
For example, if save_top_k=1, then when a new best model is found, the checkpoint will overwrite on the hard drive.

But what if I want to reduce the load on the hard drive and save the final best model only ONE time at the very end of training?

The parameter save_last=True will not work, because the last state of the model will not be optimal (I am using EarlyStopping with patience=100).