diff --git a/paddle/gserver/gradientmachines/NeuralNetwork.cpp b/paddle/gserver/gradientmachines/NeuralNetwork.cpp
index dcf0acb5a2cc7698625a2e254d4c12a32bc9631d..dbadc352a4ccd7483bf67e1025c212f514e32a24 100644
--- a/paddle/gserver/gradientmachines/NeuralNetwork.cpp
+++ b/paddle/gserver/gradientmachines/NeuralNetwork.cpp
@@ -21,6 +21,10 @@ limitations under the License. */
 #include "paddle/utils/Logging.h"
 #include "paddle/utils/Stat.h"
 
+#ifdef PADDLE_USE_MKLDNN
+#include "paddle/gserver/layers/MKLDNNLayer.h"
+#endif
+
 #ifndef PADDLE_MOBILE_INFERENCE
 #include "MultiNetwork.h"
 #include "RecurrentGradientMachine.h"
@@ -300,6 +304,17 @@ void NeuralNetwork::backward(const UpdateCallback& callback) {
   }
 }
 
+void NeuralNetwork::finish() {
+#ifdef PADDLE_USE_MKLDNN
+  FOR_EACH_R(layer, layers_) {
+    MKLDNNLayerPtr dnnLayer = std::dynamic_pointer_cast<MKLDNNLayer>(*layer);
+    if (dnnLayer) {
+      dnnLayer->convertWeightsToPaddle();
+    }
+  }
+#endif
+}
+
 Argument NeuralNetwork::getLayerOutput(const std::string& layerName) {
   return getLayer(layerName)->getOutput();
 }
diff --git a/paddle/gserver/gradientmachines/NeuralNetwork.h b/paddle/gserver/gradientmachines/NeuralNetwork.h
index 56a1ec78460731554c9b47cf3f517f7654dc314f..16971883b42786e2cb48faafd2a7e95d45075ac8 100644
--- a/paddle/gserver/gradientmachines/NeuralNetwork.h
+++ b/paddle/gserver/gradientmachines/NeuralNetwork.h
@@ -134,6 +134,9 @@ public:
 
   const std::string& getName() const { return subModelName_; }
 
+  /// some finish work, like convert the weight format of MKLDNNLayers
+  void finish() override;
+
 protected:
   /**
    * The constructor of NeuralNetwork.