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