@@ -197,2 +213,4 @@
     #     device = torch.device("cpu")
     # define loss function (criterion), optimizer, and learning rate scheduler
+
+    if not torch.cuda.is_available() and not torch.backends.mps.is_available():

@@ -126,2 +125,3 @@
     #     main_worker(args.gpu, ngpus_per_node, args)
-    main_worker(args.gpu, 1, args)
+
+    args.distributed = args.world_size > 1 or args.multiprocessing_distributed

@@ -475,3 +518,1 @@
-    #     dist.all_reduce(total, dist.ReduceOp.SUM, async_op=False)
-    #     self.sum, self.count = total.tolist()
-    #     self.avg = self.sum / self.count
+    def all_reduce(self):

@@ -10,3 +10,3 @@
 cd ./device
 echo "start data parallel training"
-mpirun -n 4 --output-filename log_output --merge-stderr-to-stdout python ./data_dp_main.py ./ImageNet2012 -a resnet50 -b 64 -dist > train.log 2>&1 &
+# mpirun -n 4 --output-filename log_output --merge-stderr-to-stdout python ./data_dp_main.py ./ImageNet2012 -a resnet50 -b 64 -dist > train.log 2>&1 &