diff --git a/apex/examples/imagenet/main_amp.py b/apex/examples/imagenet/main_amp.py
index bddf24827e..2591992c35 100644
--- a/apex/examples/imagenet/main_amp.py
+++ b/apex/examples/imagenet/main_amp.py
@@ -140,11 +140,10 @@ def main():
     model = model.cuda()
 
     # Scale learning rate based on global batch size
-    args.lr = args.lr*float(args.batch_size*args.world_size)/256. 
+    args.lr = args.lr * float(args.batch_size * args.world_size) / 256.
     optimizer = torch.optim.SGD(model.parameters(), args.lr,
                                 momentum=args.momentum,
                                 weight_decay=args.weight_decay)
-
     # Initialize Amp.  Amp accepts either values or strings for the optional override arguments,
     # for convenient interoperation with argparse.
     model, optimizer = amp.initialize(model, optimizer,