diff --git a/tests/st/ge/dynamic_shape/test_dynamic_shape_input_train.py b/tests/st/ge/dynamic_shape/test_dynamic_shape_input_train.py
index 867d6e7c296..d05bda21d75 100644
--- a/tests/st/ge/dynamic_shape/test_dynamic_shape_input_train.py
+++ b/tests/st/ge/dynamic_shape/test_dynamic_shape_input_train.py
@@ -147,13 +147,13 @@ def create_dataset(batch_size=32):
 class LossCallBack(LossMonitor):
     def __init__(self):
         super(LossCallBack, self).__init__()
-        self.last_5_losses = []
+        self.last_10_losses = []
 
     def step_end(self, run_context):
         cb_params = run_context.original_args()
         loss = cb_params.net_outputs
         loss = np.mean(loss.asnumpy())
-        self.last_5_losses = self.last_5_losses[-4:] + [loss]
+        self.last_10_losses = self.last_10_losses[-9:] + [loss]
 
 
 def train(batch_size, lr, momentum, epochs, dataset_sink_mode):
@@ -173,11 +173,11 @@ def train(batch_size, lr, momentum, epochs, dataset_sink_mode):
     model.train(epochs, dummy_dataset, callbacks=[loss_callback],
                 sink_size=dummy_dataset.get_dataset_size(), dataset_sink_mode=dataset_sink_mode)
 
-    avg_loss = np.mean(loss_callback.last_5_losses)
+    avg_loss = np.min(loss_callback.last_10_losses)
     return avg_loss
 
 
-@pytest.mark.level1
+@pytest.mark.level0
 @pytest.mark.platform_arm_ascend_training
 @pytest.mark.platform_x86_ascend_training
 @pytest.mark.env_onecard