ddp trainer doesn't work for vision transformer models

[melihcatal]

CNN models (i.e., resnet18,50 etc.) work well. vision transformer models (i.e., swin_v2_s) train on a single GPU; however, DDP training fails.