DDP evaluation / tensorboard logging

Torch.distributed.barrier() hangs in DDP - #6 by Manuel_Alejandro_Dia okay calling self.model.module helps with getting the test evaluation. I will now also check whether it solves the tensorboard issue.