Random weights are provided for the ViT model. During each step, the model selects a random subset of the masked image patches.
-