diff --git a/.gitignore b/.gitignore
index 547e94ea11f048c9e007996d4ee716d22a13742e..4c86068b7ee3024416094613d2f8f8d74ce89921 100644
--- a/.gitignore
+++ b/.gitignore
@@ -75,4 +75,5 @@ cmake-build-release
 demo/ios/PaddleMobileDemo/PaddleMobileDemo/googlenet_combine/
 demo/ios/PaddleMobileDemo/PaddleMobileDemo/*.jpg
 demo/ios/PaddleMobileDemo/PaddleMobileDemo/PaddleMobile/*.a
-*.xcuserstate
\ No newline at end of file
+*.xcuserstate
+/tools/quantification/quantify
diff --git a/src/io/loader.cpp b/src/io/loader.cpp
index 9ed877d05d51dfbe7139ea2289fdb6480c62f88f..cdcecf02ab8af22dec0e32113052ac26e9c5fcfc 100644
--- a/src/io/loader.cpp
+++ b/src/io/loader.cpp
@@ -56,7 +56,8 @@ template <typename Dtype, Precision P>
 const framework::Program<Dtype, P> Loader<Dtype, P>::Load(
     const std::string &model_path, const std::string &para_path, bool optimize,
     bool quantification) {
-  auto program = this->LoadProgram(model_path, optimize);
+  auto program = this->LoadProgram(model_path, optimize, quantification);
+
   program.para_path = para_path;
   program.combined = true;
   program.quantification = quantification;
diff --git a/src/jni/paddle_mobile_jni.cpp b/src/jni/paddle_mobile_jni.cpp
index d4eb9e0f0733814cbe367a1873e241383127340a..66150e24e0ac957773161904948c10cf4637ee42 100644
--- a/src/jni/paddle_mobile_jni.cpp
+++ b/src/jni/paddle_mobile_jni.cpp
@@ -61,6 +61,15 @@ JNIEXPORT jboolean JNICALL Java_com_baidu_paddle_PML_load(JNIEnv *env,
                                          optimize);
 }
 
+JNIEXPORT jboolean JNICALL Java_com_baidu_paddle_PML_loadQualified(
+    JNIEnv *env, jclass thiz, jstring modelPath) {
+  ANDROIDLOGI("loadQualified invoked");
+  bool optimize = true;
+  bool qualified = true;
+  return getPaddleMobileInstance()->Load(jstring2cppstring(env, modelPath),
+                                         optimize, qualified);
+}
+
 JNIEXPORT jboolean JNICALL Java_com_baidu_paddle_PML_loadCombined(
     JNIEnv *env, jclass thiz, jstring modelPath, jstring paramPath) {
   ANDROIDLOGI("loadCombined invoked");
@@ -70,6 +79,16 @@ JNIEXPORT jboolean JNICALL Java_com_baidu_paddle_PML_loadCombined(
                                          optimize);
 }
 
+JNIEXPORT jboolean JNICALL Java_com_baidu_paddle_PML_loadCombinedQualified(
+    JNIEnv *env, jclass thiz, jstring modelPath, jstring paramPath) {
+  ANDROIDLOGI("loadCombinedQualified invoked");
+  bool optimize = true;
+  bool qualified = true;
+  return getPaddleMobileInstance()->Load(jstring2cppstring(env, modelPath),
+                                         jstring2cppstring(env, paramPath),
+                                         optimize, qualified);
+}
+
 JNIEXPORT jfloatArray JNICALL Java_com_baidu_paddle_PML_predictImage(
     JNIEnv *env, jclass thiz, jfloatArray buf, jintArray ddims) {
   ANDROIDLOGI("predictImage invoked");
diff --git a/src/jni/paddle_mobile_jni.h b/src/jni/paddle_mobile_jni.h
index a830ab43c8ee0598fbf75e1fef5f3eb7da06c27b..06fabe04c739dfcee06110a3592a88591e3d37b9 100644
--- a/src/jni/paddle_mobile_jni.h
+++ b/src/jni/paddle_mobile_jni.h
@@ -27,12 +27,24 @@ namespace jni {
 JNIEXPORT jboolean JNICALL Java_com_baidu_paddle_PML_load(JNIEnv *env,
                                                           jclass thiz,
                                                           jstring modelPath);
+
+/**
+ * load separated qualified model for android
+ */
+JNIEXPORT jboolean JNICALL Java_com_baidu_paddle_PML_loadQualified(
+    JNIEnv *env, jclass thiz, jstring modelPath);
 /**
  * load combined model  for android
  */
 JNIEXPORT jboolean JNICALL Java_com_baidu_paddle_PML_loadCombined(
     JNIEnv *env, jclass thiz, jstring modelPath, jstring paramPath);
 
+/**
+ * load combined qualified model for android
+ */
+JNIEXPORT jboolean JNICALL Java_com_baidu_paddle_PML_loadCombinedQualified(
+    JNIEnv *env, jclass thiz, jstring modelPath, jstring paramPath);
+
 /**
  * object detection for anroid
  */
diff --git a/tools/quantification/convert.cpp b/tools/quantification/convert.cpp
index 0223a3b353c4715ed0c52a1d6b3f0751e372691d..282b22073fc96ddb2ed0d421f113604aadcc4afc 100644
--- a/tools/quantification/convert.cpp
+++ b/tools/quantification/convert.cpp
@@ -68,60 +68,60 @@ std::shared_ptr<ProgramDesc> loadParams(const std::string &model_path) {
 
 }
 
-void LoadWithDump(const paddle_mobile::framework::VarDesc &var_desc, char *dataP, FILE *out_file) {
+void LoadWithDump(const paddle_mobile::framework::VarDesc &var_desc, char **dataP, FILE *out_file) {
     // 1. version
-    uint32_t version = *reinterpret_cast<uint32_t *>(dataP);
+    uint32_t version = *reinterpret_cast<uint32_t *>(*dataP);
 
     // write version
     fwrite(&version, kSize32, 1, out_file);
 
-    dataP += kSize32;
+    *dataP += kSize32;
 
     // 2 Lod information
     auto *lod_level_ptr = new uint64_t();
-    memcpy(lod_level_ptr, dataP, kSize64);
+    memcpy(lod_level_ptr, *dataP, kSize64);
 
     uint64_t lod_level = 0;
     // write lod Information
     fwrite(&lod_level, kSize64, 1, out_file);
     delete lod_level_ptr;
 
-    dataP += kSize64;
+    *dataP += kSize64;
 
     for (uint64_t i = 0; i < lod_level; ++i) {
-        uint64_t size = *reinterpret_cast<uint64_t *>(dataP);
+        uint64_t size = *reinterpret_cast<uint64_t *>(*dataP);
         // write lod size
         fwrite(&size, kSize64, 1, out_file);
-        (dataP) += kSize64;
+        (*dataP) += kSize64;
 
         std::vector<size_t> tmp(size / sizeof(size_t));
         for (unsigned long &k : tmp) {
-            k = *reinterpret_cast<size_t *>(dataP);
-            (dataP) += sizeof(size_t);
+            k = *reinterpret_cast<size_t *>(*dataP);
+            (*dataP) += sizeof(size_t);
         }
         // write lod size vector
         fwrite(&tmp, sizeof(size_t), tmp.size(), out_file);
     }
 
     // 3. tensor version
-    uint32_t tensor_version = *reinterpret_cast<uint32_t *>(dataP);
+    uint32_t tensor_version = *reinterpret_cast<uint32_t *>(*dataP);
     // write tensor version
     fwrite(&tensor_version, kSize32, 1, out_file);
-    (dataP) += kSize32;
+    (*dataP) += kSize32;
 
     // 4. tensor desc
-    int32_t size = *reinterpret_cast<int32_t *>(dataP);
+    int32_t size = *reinterpret_cast<int32_t *>(*dataP);
     // write tensor desc
     fwrite(&size, sizeof(int32_t), 1, out_file);
-    (dataP) += sizeof(int32_t);
+    (*dataP) += sizeof(int32_t);
 
     std::unique_ptr<char[]> buf(new char[size]);
     for (int m = 0; m < size; ++m) {
-        buf.get()[m] = (dataP)[m];
+        buf.get()[m] = (*dataP)[m];
     }
 
     fwrite(buf.get(), sizeof(char), static_cast<size_t>(size), out_file);
-    (dataP) += (sizeof(char) * size);
+    (*dataP) += (sizeof(char) * size);
 
     const paddle_mobile::framework::TensorDesc &desc = var_desc.Tensor_desc();
     int memory_size = 1;
@@ -158,9 +158,9 @@ void LoadWithDump(const paddle_mobile::framework::VarDesc &var_desc, char *dataP
     memory = new char[tensorSize];
 
     for (int n = 0; n < tensorSize; ++n) {
-        static_cast<char *>(memory)[n] = (dataP)[n];
+        static_cast<char *>(memory)[n] = (*dataP)[n];
     }
-    dataP += tensorSize;
+    *dataP += tensorSize;
 
     // for float 32
     float min_value = std::numeric_limits<float>::max();
@@ -194,7 +194,7 @@ quantificate_combined(const std::string &model_path, const std::string &param_pa
                 if (var_desc->Name() == "feed" || var_desc->Name() == "fetch") {
                     continue;
                 }
-                LoadWithDump(*var_desc, data, out_file);
+                LoadWithDump(*var_desc, &data, out_file);
             }
         }
     }
@@ -220,7 +220,7 @@ void quantificate_seperated(const std::string model_dir, const std::string param
                 FILE *out_file = fopen(file_name.c_str(), "wb");
                 char *origin_data = Get_binary_data(model_dir + "/" + var_desc->Name());
                 char *data = origin_data;
-                LoadWithDump(*var_desc, data, out_file);
+                LoadWithDump(*var_desc, &data, out_file);
                 delete origin_data;
                 fclose(out_file);
             }