we presently run inference on gpus, we train on tpus

→ View original post on X — @midjourney