KeqiangSun

Hi, I try to train this modoel with 12GB GPUs, with "--gpus=8 --batch=8 --mbstd-group=1". But it stuck and I noticed the memory on the first GPU exhausted at a moment....