--cache_for_epochs with augmentation

[hhiyassat]

in training new model and using augmentation, does the usage of --cache_for_epochs leads to model over-fitting
should I eliminate --cache_for_epochs when using augmentation?