Saving model (Based on accuracy)

Hi all,

Working with this codebase it looks like it doesn't check verification (or any form of) accuracy before saving the model like most deep learning implementation. Why is that? Isn't there a possibility of overfitting?

I feel like model that doesn't perform well as the previous ones might be saved thereby leading to a poorer performance when validation on a different dataset.

Please clarify/explain.

Thanks