diff --git a/paddlepalm/task_paradigm/match.py b/paddlepalm/task_paradigm/match.py
index d42c64a7d32cdf3aa3d8fb32ed7890c6f41119c5..df4e43bc33158fd3526e3b8caea61106ed8db439 100644
--- a/paddlepalm/task_paradigm/match.py
+++ b/paddlepalm/task_paradigm/match.py
@@ -79,8 +79,8 @@ class TaskParadigm(task_paradigm):
                 initializer=fluid.initializer.Constant(0.)))
 
         if self._is_training:
-            ce_loss, probs = fluid.layers.cross_entropy(
-                input=logits, label=labels, return_softmax=True)
+            ce_loss = fluid.layers.cross_entropy(
+                input=logits, label=labels)
             loss = fluid.layers.mean(x=ce_loss)
             return {'loss': loss}
         else: