papersnake·il y a 4 ans·discussHave you tested this on big models involving multi-gpu communication, or any plans?