7 ms·compute and improvements in pretraning and RLby in_making 7d agocompute and improvements in pretraning and RL