diff --git a/tests/st/heterogeneous/test_mem_scheduler.py b/tests/st/heterogeneous/test_mem_scheduler.py
index 9eb7e3dbd27..f0c34b76661 100644
--- a/tests/st/heterogeneous/test_mem_scheduler.py
+++ b/tests/st/heterogeneous/test_mem_scheduler.py
@@ -154,10 +154,9 @@ def test_lenet_manual_offload():
     os.environ['ENABLE_MEM_SCHEDULER'] = '0'
 
 
-@pytest.mark.level1
+@pytest.mark.level0
 @pytest.mark.platform_arm_ascend_training
 @pytest.mark.platform_x86_ascend_training
-@pytest.mark.platform_x86_gpu_training
 @pytest.mark.env_onecard
 def test_1024_batch_size_resnet():
     """
@@ -167,7 +166,7 @@ def test_1024_batch_size_resnet():
     """
     os.environ['GRAPH_OP_RUN'] = '1'
     num_classes = 10
-    epoch = 4
+    epoch = 6
     batch_size = 1024
     context.set_context(memory_offload='ON')
     net = resnet50(batch_size, num_classes)
@@ -177,8 +176,7 @@ def test_1024_batch_size_resnet():
                                 net.get_parameters()), lr, momentum)
     criterion = nn.SoftmaxCrossEntropyWithLogits(sparse=True, reduction='mean')
     net_with_criterion = WithLossCell(net, criterion)
-    train_network = TrainOneStepCell(
-        net_with_criterion, optimizer)  # optimizer
+    train_network = TrainOneStepCell(net_with_criterion, optimizer)  # optimizer
     train_network.set_train()
     losses = []
     for _ in range(0, epoch):
@@ -187,5 +185,5 @@ def test_1024_batch_size_resnet():
         label = Tensor(np.ones([batch_size]).astype(np.int32))
         loss = train_network(data, label)
         losses.append(loss)
-    assert losses[-1].asnumpy() < 1
+    assert losses[-1].asnumpy() < 1.5
     os.environ['GRAPH_OP_RUN'] = '0'