Pretrain with 40G A100 or 80G A100?
Closed this issue · 1 comments
basaltzhang commented
As title
LeeYN-43 commented
Sorry for the late reply. 40G A100 is fine. When each card has a batch size of 16, it will take up about 31G memory.
Closed this issue · 1 comments
As title
Sorry for the late reply. 40G A100 is fine. When each card has a batch size of 16, it will take up about 31G memory.