Hello author, thank you very much for your work, it has been very inspiring for us. I would like to ask about Table 8 in the paper, for the w/o human column, did you train all other modules from scratch on the Robotwin2.0 single-task dataset except for loading the dino-siglip weights?