diff --git a/paddle/framework/executor.cc b/paddle/framework/executor.cc
index 03504952edc22288191cb1558956e89819348d81..aa36b7438ff8e34fdd5435f8b330046a407eed26 100644
--- a/paddle/framework/executor.cc
+++ b/paddle/framework/executor.cc
@@ -27,12 +27,12 @@ Executor::Executor(const std::vector<platform::Place>& places) {
   device_contexts_.resize(places.size());
   for (size_t i = 0; i < places.size(); i++) {
     if (platform::is_cpu_place(places[i])) {
-      device_contexts_[i].reset(new platform::CPUDeviceContext(
-          boost::get<platform::CPUPlace>(places[i])));
-    } else {
+      device_contexts_[i] = new platform::CPUDeviceContext(
+          boost::get<platform::CPUPlace>(places[i]));
+    } else if (platform::is_gpu_place(places[i])) {
 #ifndef PADDLE_ONLY_CPU
-      device_contexts_[i].reset(new platform::CUDADeviceContext(
-          boost::get<platform::CPUPlace>(places[i])));
+      device_contexts_[i] = new platform::CUDADeviceContext(
+          boost::get<platform::GPUPlace>(places[i]));
 #else
       PADDLE_THROW("'GPUPlace' is not supported in CPU only device.");
 #endif
@@ -40,6 +40,14 @@ Executor::Executor(const std::vector<platform::Place>& places) {
   }
 }
 
+Executor::~Executor() {
+  for (auto& device_context : device_contexts_) {
+    if (device_context) {
+      delete device_context;
+    }
+  }
+}
+
 void Executor::Run(const ProgramDesc& pdesc, Scope* scope,
                    std::vector<Tensor>* outputs) {
   // TODO(tonyyang-svail):
@@ -59,6 +67,7 @@ void Executor::Run(const ProgramDesc& pdesc, Scope* scope,
 
   for (auto& op_desc : block.ops()) {
     auto op = paddle::framework::OpRegistry::CreateOp(op_desc);
+    std::cout << op->DebugString() << std::endl;
     op->Run(*scope, *device);
   }
 
diff --git a/paddle/framework/executor.h b/paddle/framework/executor.h
index d5c21c59feb5ad3b238348ae5b343eddf5c3f4ed..fc53be37c3f2806f635d1d32559e86d8f65998db 100644
--- a/paddle/framework/executor.h
+++ b/paddle/framework/executor.h
@@ -25,11 +25,11 @@ namespace framework {
 class Executor {
  public:
   explicit Executor(const std::vector<platform::Place>& places);
-  ~Executor() {}
+  ~Executor();
   void Run(const ProgramDesc&, Scope*, std::vector<Tensor>*);
 
  private:
-  std::vector<std::unique_ptr<platform::DeviceContext>> device_contexts_;
+  std::vector<platform::DeviceContext*> device_contexts_;
 };
 
 }  // namespace framework
diff --git a/paddle/platform/gpu_info.cc b/paddle/platform/gpu_info.cc
index f48701487195a28450467542f2167b07a3a93294..0464797f3107f4d6ac7ae7d7b087a955334c82e4 100644
--- a/paddle/platform/gpu_info.cc
+++ b/paddle/platform/gpu_info.cc
@@ -43,7 +43,7 @@ int GetCurrentDeviceId() {
 }
 
 void SetDeviceId(int id) {
-  PADDLE_ENFORCE(id < GetDeviceCount(), "id must less than GPU count")
+  PADDLE_ENFORCE(id < GetDeviceCount(), "id must less than GPU count");
   PADDLE_ENFORCE(cudaSetDevice(id),
                  "cudaSetDevice failed in paddle::platform::SetDeviceId");
 }