mstar.model.command_a_plus.components# Modules vision Command A+'s image encoder: a SigLIP tower, a pixel shuffle and a SwiGLU projector.