diff --git a/examples/mxnet/gat/gat_batch.py b/examples/mxnet/gat/gat_batch.py
index ecd36e31a4a7..44545ced910d 100644
--- a/examples/mxnet/gat/gat_batch.py
+++ b/examples/mxnet/gat/gat_batch.py
@@ -178,10 +178,11 @@ def main(args):
         test_mask = test_mask.as_in_context(ctx)
         val_mask = val_mask.as_in_context(ctx)
     # create graph
-    g = DGLGraph(data.graph)
+    g = data.graph
     # add self-loop
+    g.remove_edges_from(g.selfloop_edges())
+    g = DGLGraph(g)
     g.add_edges(g.nodes(), g.nodes())
-
     # create model
     model = GAT(g,
                 args.num_layers,
diff --git a/examples/mxnet/gcn/train.py b/examples/mxnet/gcn/train.py
index dd574d9150a1..1c90099bee3f 100644
--- a/examples/mxnet/gcn/train.py
+++ b/examples/mxnet/gcn/train.py
@@ -52,9 +52,11 @@ def main(args):
     test_mask = test_mask.as_in_context(ctx)
 
     # create GCN model
-    g = DGLGraph(data.graph)
+    g = data.graph
     if args.self_loop:
-        g.add_edges(g.nodes(), g.nodes())
+        g.remove_edges_from(g.selfloop_edges())
+        g.add_edges_from(zip(g.nodes(), g.nodes()))
+    g = DGLGraph(g)
     # normalization
     degs = g.in_degrees().astype('float32')
     norm = mx.nd.power(degs, -0.5)
diff --git a/examples/pytorch/gat/train.py b/examples/pytorch/gat/train.py
index 5ff3d9efca3c..e45c334074e1 100644
--- a/examples/pytorch/gat/train.py
+++ b/examples/pytorch/gat/train.py
@@ -65,11 +65,12 @@ def main(args):
         val_mask = val_mask.cuda()
         test_mask = test_mask.cuda()
 
-    # create DGL graph
-    g = DGLGraph(data.graph)
-    n_edges = g.number_of_edges()
+    g = data.graph
     # add self loop
+    g.remove_edges_from(g.selfloop_edges())
+    g = DGLGraph(g)
     g.add_edges(g.nodes(), g.nodes())
+    n_edges = g.number_of_edges()
     # create model
     heads = ([args.num_heads] * args.num_layers) + [args.num_out_heads]
     model = GAT(g,
diff --git a/examples/pytorch/gcn/gcn_mp.py b/examples/pytorch/gcn/gcn_mp.py
index b138636a8ff3..24affad94840 100644
--- a/examples/pytorch/gcn/gcn_mp.py
+++ b/examples/pytorch/gcn/gcn_mp.py
@@ -147,10 +147,12 @@ def main(args):
         test_mask = test_mask.cuda()
 
     # graph preprocess and calculate normalization factor
-    g = DGLGraph(data.graph)
-    n_edges = g.number_of_edges()
+    g = data.graph
+    g.remove_edges_from(g.selfloop_edges())
+    g = DGLGraph(g)
     # add self loop
     g.add_edges(g.nodes(), g.nodes())
+    n_edges = g.number_of_edges()
     # normalization
     degs = g.in_degrees().float()
     norm = torch.pow(degs, -0.5)
diff --git a/examples/pytorch/gcn/train.py b/examples/pytorch/gcn/train.py
index 153679db968b..bbfeb57accfb 100644
--- a/examples/pytorch/gcn/train.py
+++ b/examples/pytorch/gcn/train.py
@@ -54,11 +54,13 @@ def main(args):
         test_mask = test_mask.cuda()
 
     # graph preprocess and calculate normalization factor
-    g = DGLGraph(data.graph)
-    n_edges = g.number_of_edges()
+    g = data.graph
     # add self loop
     if args.self_loop:
-        g.add_edges(g.nodes(), g.nodes())
+            g.remove_edges_from(g.selfloop_edges())
+            g.add_edges_from(zip(g.nodes(), g.nodes()))
+    g = DGLGraph(g)
+    n_edges = g.number_of_edges()
     # normalization
     degs = g.in_degrees().float()
     norm = torch.pow(degs, -0.5)