fix resnext50 gpu amp

model_zoo/official/cv/resnext50/train.py

@@ -240,12 +240,8 @@ def train(cloud_args=None):
     else:
         loss_scale_manager = FixedLossScaleManager(args.loss_scale, drop_overflow_update=False)
 
-    if args.platform == "Ascend":
-        model = Model(network, loss_fn=loss, optimizer=opt, loss_scale_manager=loss_scale_manager,
-                      metrics={'acc'}, amp_level="O3")
-    else:
-        model = Model(network, loss_fn=loss, optimizer=opt, loss_scale_manager=loss_scale_manager,
-                      metrics={'acc'}, amp_level="O2")
+    model = Model(network, loss_fn=loss, optimizer=opt, loss_scale_manager=loss_scale_manager,
+                  metrics={'acc'}, amp_level="O3")
 
     # checkpoint save
     progress_cb = ProgressMonitor(args)