diff --git a/CMakeLists.txt b/CMakeLists.txt
index 1abe068f0c6a2a54b4bcdcc675630859ef499c63..989f96e311ef19cd3a8722833884c15538e698ce 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -11,10 +11,10 @@ else()
     set(CMAKE_BUILD_TYPE Release)
 endif ()
 
+
 if(DEBUGING)
     message(STATUS "debuging")
     add_definitions(-DPADDLE_MOBILE_DEBUG)
-
 else()
     message(STATUS "releasing")
     add_definitions(-fvisibility=hidden -fvisibility-inlines-hidden)
@@ -25,7 +25,6 @@ if (USE_EXCEPTION)
     add_definitions(-fexceptions)
 else()
     add_definitions(-fno-exceptions)
-
 endif ()
 
 if(IS_MAC)
@@ -119,7 +118,6 @@ else ()
     add_definitions(-DTRANSPOSE_OP)
 endif()
 
-
 add_library(paddle-mobile SHARED ${PADDLE_MOBILE_CC} ${PADDLE_MOBILE_H})
 
 if(DEBUGING)
diff --git a/src/io/io.cpp b/src/io/io.cpp
index 465f28bc8338ff4217b61db8ebdb5151e0210449..b37b3e248aac6839a268121ced5c9ee438cccc05 100644
--- a/src/io/io.cpp
+++ b/src/io/io.cpp
@@ -279,17 +279,14 @@ Executor<Dtype, P>::Executor(const framework::Program<Dtype> p, int batch_size,
 
 template <typename Dtype, Precision P>
 void Executor<Dtype, P>::LoadMemory(const framework::VarDesc var_desc,
-                                    framework::LoDTensor *tensor,
-                                    const std::string &file_path, char *data) {
+                                    framework::LoDTensor *tensor, char *&data) {
   // 1. version
   uint32_t version = *(uint32_t *)data;
   data += sizeof(uint32_t);
-  DLOG << "version: " << version;
 
   // 2 Lod information
   uint64_t lod_level = *(uint64_t *)data;
   data += sizeof(uint64_t);
-  DLOG << "lod_level: " << lod_level;
 
   auto &lod = *tensor->mutable_lod();
   lod.resize(lod_level);
@@ -297,7 +294,6 @@ void Executor<Dtype, P>::LoadMemory(const framework::VarDesc var_desc,
     uint64_t size = *(uint64_t *)data;
     data += sizeof(uint64_t);
     DLOG << "lod size: " << i << size;
-
     std::vector<size_t> tmp(size / sizeof(size_t));
 
     for (int k = 0; k < tmp.size(); ++k) {
@@ -315,12 +311,10 @@ void Executor<Dtype, P>::LoadMemory(const framework::VarDesc var_desc,
   // 3. tensor version
   uint32_t tensor_version = *(uint32_t *)data;
   data += sizeof(uint32_t);
-  DLOG << "tensor_version: " << tensor_version;
 
   // 4. tensor desc
   int32_t size = *(int32_t *)data;
   data += sizeof(int32_t);
-  DLOG << "tensor desc size: " << size;
 
   std::unique_ptr<char[]> buf(new char[size]);
   for (int m = 0; m < size; ++m) {
@@ -344,7 +338,6 @@ void Executor<Dtype, P>::LoadMemory(const framework::VarDesc var_desc,
       break;
     case framework::VARTYPE_TYPE_FP32:
       type_size = 4;
-      DLOG << " type size: " << type_size;
       memory = tensor->mutable_data<float>();
       break;
     case framework::VARTYPE_TYPE_FP64:
@@ -382,8 +375,8 @@ void Executor<Dtype, P>::InitMemory() {
 
         char *origin_data =
             Get_binary_data(program_.model_path + "/" + var_desc->Name());
-        LoadMemory(*var_desc, tensor,
-                   program_.model_path + "/" + var_desc->Name(), origin_data);
+        char *data = origin_data;
+        LoadMemory(*var_desc, tensor, data);
         delete origin_data;
       } else {
         if (var_desc->Type() == framework::VARTYPE_TYPE_LOD_TENSOR) {
@@ -399,7 +392,7 @@ void Executor<Dtype, P>::InitMemory() {
 template <typename Dtype, Precision P>
 void Executor<Dtype, P>::InitCombineMemory() {
   char *origin_data = Get_binary_data(program_.para_path);
-
+  char *data = origin_data;
   for (const auto &block : to_predict_program_->Blocks()) {
     for (const auto &var_desc : block->Vars()) {
       auto var = program_.scope->Var(var_desc->Name());
@@ -408,18 +401,15 @@ void Executor<Dtype, P>::InitCombineMemory() {
         if (var_desc->Name() == "feed" || var_desc->Name() == "fetch") {
           continue;
         }
-        LoadMemory(*var_desc, tensor,
-                   program_.model_path + "/" + var_desc->Name(), origin_data);
+        LoadMemory(*var_desc, tensor, data);
       } else {
         if (var_desc->Type() == framework::VARTYPE_TYPE_LOD_TENSOR) {
           auto tensor = var->template GetMutable<framework::LoDTensor>();
-
           tensor->template mutable_data<Ptype>();
         }
       }
     }
   }
-
   delete origin_data;
 }
 
diff --git a/src/io/io.h b/src/io/io.h
index f89ccc8b2efeb755d3386e224ee61557642e03e9..2eaa94d0a63eadfc1a41761a71b2ea31a0fc7b07 100644
--- a/src/io/io.h
+++ b/src/io/io.h
@@ -63,8 +63,7 @@ class Executor {
 
   void InitMemory();
   void LoadMemory(const framework::VarDesc var_desc,
-                  framework::LoDTensor *tensor, const std::string &file_path,
-                  char *data);
+                  framework::LoDTensor *tensor, char *&data);
   void InitCombineMemory();
   framework::Program<Dtype> program_;
   int batch_size_ = 1;
diff --git a/test/net/test_googlenet.cpp b/test/net/test_googlenet.cpp
index 210f9643a165909df9bbade5b842c7af9840cce3..ab4fd2fe0d1eaaa58fabc38fbf512a0b860c36f0 100644
--- a/test/net/test_googlenet.cpp
+++ b/test/net/test_googlenet.cpp
@@ -20,10 +20,9 @@ int main() {
   paddle_mobile::Loader<paddle_mobile::CPU> loader;
   bool optimize = false;
   auto time1 = time();
-
-  auto program = loader.Load(g_googlenet, optimize);
-  //  auto program = loader.Load(g_googlenet_combine + "/model",
-  //          g_googlenet_combine + "/params", optimize);
+  //  auto program = loader.Load(g_googlenet, optimize);
+  auto program = loader.Load(g_googlenet_combine + "/model",
+                             g_googlenet_combine + "/params", optimize);
   auto time2 = time();
   DLOG << "load cost :" << time_diff(time1, time2) << "ms\n";
   paddle_mobile::Executor<paddle_mobile::CPU> executor(program, 1, optimize);