papersnake·4 yıl önce·discussHave you tested this on big models involving multi-gpu communication, or any plans?