diff --git a/paddle/platform/device_context.h b/paddle/platform/device_context.h
index a106592e454e21c46cd2f87f1bbf6694955d6e23..f6a39a8e26c301296aac0af7f4e8b2c6c97ece24 100644
--- a/paddle/platform/device_context.h
+++ b/paddle/platform/device_context.h
@@ -34,13 +34,14 @@ class DeviceContext {
 
   template <typename DeviceType>
   DeviceType* get_eigen_device() const;
+
+  virtual void Wait() const {}
 };
 
 class CPUDeviceContext : public DeviceContext {
  public:
   CPUDeviceContext();
   explicit CPUDeviceContext(CPUPlace place);
-  virtual ~CPUDeviceContext() {}
 
   Eigen::DefaultDevice* eigen_device() const;
 
@@ -59,7 +60,7 @@ class CUDADeviceContext : public DeviceContext {
   virtual ~CUDADeviceContext();
 
   /*! \brief  Wait for all operations completion in the stream. */
-  void Wait() const;
+  void Wait() const override;
 
   /*! \brief  Return place in the device context. */
   Place GetPlace() const override;
diff --git a/paddle/pybind/pybind.cc b/paddle/pybind/pybind.cc
index c7009a604f60cda11434ad33b6c7d7caee1befdd..f5cba1b78fbed3a6f7fa57db46e396eb10bfed6e 100644
--- a/paddle/pybind/pybind.cc
+++ b/paddle/pybind/pybind.cc
@@ -238,7 +238,13 @@ All parameter, weight, gradient are variables in Paddle.
              return Backward(forwardOp, no_grad_vars).release();
            })
       .def("infer_shape", &OperatorBase::InferShape)
-      .def("run", &OperatorBase::Run)
+      .def("run",
+           [](OperatorBase &self,
+              const Scope &scope,
+              const platform::DeviceContext &dev_ctx) {
+             self.Run(scope, dev_ctx);
+             dev_ctx.Wait();
+           })
       .def("type",
            [](const OperatorBase &op) -> std::string { return op.Type(); })
       .def("outputs",