Why do I use distributed training fedSEG? The usage of each GPU is very inconsistent. Some Gpus even run out of memory, while others do not.
FedML-AI
