print model parameters

By integrating contrastive, self-supervised, and reconstruction learning, we have trained numerous visual tokenizers from scratch. We are se