Exascale Training of Generative Model with Historical Priors for Data Reduction