diff --git a/mindspore/lite/examples/quick_start_cpp/CMakeLists.txt b/mindspore/lite/examples/quick_start_cpp/CMakeLists.txt
new file mode 100644
index 0000000000..612e9288c4
--- /dev/null
+++ b/mindspore/lite/examples/quick_start_cpp/CMakeLists.txt
@@ -0,0 +1,21 @@
+cmake_minimum_required(VERSION 3.14)
+project(QuickStartCpp)
+
+if(CMAKE_CXX_COMPILER_ID STREQUAL "GNU" AND CMAKE_CXX_COMPILER_VERSION VERSION_LESS 7.3.0)
+    message(FATAL_ERROR "GCC version ${CMAKE_CXX_COMPILER_VERSION} must not be less than 7.3.0")
+endif()
+
+include_directories(${CMAKE_CURRENT_SOURCE_DIR})
+file(GLOB_RECURSE QUICK_START_CXX ${CMAKE_CURRENT_SOURCE_DIR}/*.cc)
+
+add_library(mindspore-lite STATIC IMPORTED)
+set_target_properties(mindspore-lite PROPERTIES IMPORTED_LOCATION
+        ${CMAKE_CURRENT_SOURCE_DIR}/lib/libmindspore-lite.a)
+
+add_executable(mindspore_quick_start_cpp ${QUICK_START_CXX})
+
+target_link_libraries(
+        mindspore_quick_start_cpp
+        -Wl,--whole-archive mindspore-lite -Wl,--no-whole-archive
+        pthread
+)
\ No newline at end of file
diff --git a/mindspore/lite/examples/quick_start_cpp/build.sh b/mindspore/lite/examples/quick_start_cpp/build.sh
new file mode 100644
index 0000000000..339782a33f
--- /dev/null
+++ b/mindspore/lite/examples/quick_start_cpp/build.sh
@@ -0,0 +1,43 @@
+#!/bin/bash
+# Copyright 2021 Huawei Technologies Co., Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ============================================================================
+
+BASEPATH=$(cd "$(dirname $0)"; pwd)
+get_version() {
+    VERSION_MAJOR=$(grep "const int ms_version_major =" ${BASEPATH}/../../include/version.h | tr -dc "[0-9]")
+    VERSION_MINOR=$(grep "const int ms_version_minor =" ${BASEPATH}/../../include/version.h | tr -dc "[0-9]")
+    VERSION_REVISION=$(grep "const int ms_version_revision =" ${BASEPATH}/../../include/version.h | tr -dc "[0-9]")
+    VERSION_STR=${VERSION_MAJOR}.${VERSION_MINOR}.${VERSION_REVISION}
+}
+get_version
+MODEL_DOWNLOAD_URL="https://download.mindspore.cn/model_zoo/official/lite/mobilenetv2_openimage_lite/1.1/mobilenetv2.ms"
+MINDSPORE_LITE_DOWNLOAD_URL="https://ms-release.obs.cn-north-4.myhuaweicloud.com/${VERSION_STR}/MindSpore/lite/release/linux/mindspore-lite-${VERSION_STR}-inference-linux-x64.tar.gz"
+
+mkdir -p build
+mkdir -p lib
+mkdir -p model
+if [ ! -e ${BASEPATH}/model/mobilenetv2.ms ]; then
+    wget -c -O ${BASEPATH}/model/mobilenetv2.ms --no-check-certificate ${MODEL_DOWNLOAD_URL}
+fi
+if [ ! -e ${BASEPATH}/build/mindspore-lite-${VERSION_STR}-inference-linux-x64.tar.gz ]; then
+    wget -c -O ${BASEPATH}/build/mindspore-lite-${VERSION_STR}-inference-linux-x64.tar.gz --no-check-certificate ${MINDSPORE_LITE_DOWNLOAD_URL}
+fi
+tar xzvf ${BASEPATH}/build/mindspore-lite-${VERSION_STR}-inference-linux-x64.tar.gz -C ${BASEPATH}/build/
+cp -r ${BASEPATH}/build/mindspore-lite-${VERSION_STR}-inference-linux-x64-avx/lib/libmindspore-lite.a ${BASEPATH}/lib
+cp -r ${BASEPATH}/build/mindspore-lite-${VERSION_STR}-inference-linux-x64-avx/include ${BASEPATH}/
+
+cd ${BASEPATH}/build
+cmake ${BASEPATH}
+make
\ No newline at end of file
diff --git a/mindspore/lite/examples/quick_start_cpp/main.cc b/mindspore/lite/examples/quick_start_cpp/main.cc
new file mode 100644
index 0000000000..802dfc7740
--- /dev/null
+++ b/mindspore/lite/examples/quick_start_cpp/main.cc
@@ -0,0 +1,179 @@
+/**
+ * Copyright 2021 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include <algorithm>
+#include <random>
+#include <iostream>
+#include <fstream>
+#include <cstring>
+#include "include/errorcode.h"
+#include "include/model.h"
+#include "include/context.h"
+#include "include/lite_session.h"
+
+char *ReadFile(const char *file, size_t *size) {
+  if (file == nullptr) {
+    std::cerr << "file is nullptr." << std::endl;
+    return nullptr;
+  }
+
+  std::ifstream ifs(file);
+  if (!ifs.good()) {
+    std::cerr << "file: " << file << " is not exist." << std::endl;
+    return nullptr;
+  }
+
+  if (!ifs.is_open()) {
+    std::cerr << "file: " << file << " open failed." << std::endl;
+    return nullptr;
+  }
+
+  ifs.seekg(0, std::ios::end);
+  *size = ifs.tellg();
+  std::unique_ptr<char[]> buf(new (std::nothrow) char[*size]);
+  if (buf == nullptr) {
+    std::cerr << "malloc buf failed, file: " << file << std::endl;
+    ifs.close();
+    return nullptr;
+  }
+
+  ifs.seekg(0, std::ios::beg);
+  ifs.read(buf.get(), *size);
+  ifs.close();
+
+  return buf.release();
+}
+
+template <typename T, typename Distribution>
+void GenerateRandomData(int size, void *data, Distribution distribution) {
+  std::mt19937 random_engine;
+  int elements_num = size / sizeof(T);
+  (void)std::generate_n(static_cast<T *>(data), elements_num,
+                        [&]() { return static_cast<T>(distribution(random_engine)); });
+}
+
+int GenerateInputDataWithRandom(std::vector<mindspore::tensor::MSTensor *> inputs) {
+  for (auto tensor : inputs) {
+    auto input_data = tensor->MutableData();
+    void *random_data = malloc(tensor->Size());
+    if (input_data == nullptr) {
+      std::cerr << "MallocData for inTensor failed." << std::endl;
+      return -1;
+    }
+    GenerateRandomData<float>(tensor->Size(), random_data, std::uniform_real_distribution<float>(0.1f, 1.0f));
+    // Copy data to input tensor.
+    memcpy(input_data, random_data, tensor->Size());
+  }
+  return mindspore::lite::RET_OK;
+}
+
+int Run(mindspore::session::LiteSession *session) {
+  auto inputs = session->GetInputs();
+  auto ret = GenerateInputDataWithRandom(inputs);
+  if (ret != mindspore::lite::RET_OK) {
+    std::cerr << "Generate Random Input Data failed." << std::endl;
+    return ret;
+  }
+
+  ret = session->RunGraph();
+  if (ret != mindspore::lite::RET_OK) {
+    std::cerr << "Inference error " << ret << std::endl;
+    return ret;
+  }
+
+  auto out_tensors = session->GetOutputs();
+  for (auto tensor : out_tensors) {
+    std::cout << "tensor name is:" << tensor.first << " tensor size is:" << tensor.second->Size()
+              << " tensor elements num is:" << tensor.second->ElementsNum() << std::endl;
+    auto out_data = reinterpret_cast<float *>(tensor.second->MutableData());
+    std::cout << "output data is:";
+    for (int i = 0; i < tensor.second->ElementsNum() && i <= 50; i++) {
+      std::cout << out_data[i] << " ";
+    }
+    std::cout << std::endl;
+  }
+  return mindspore::lite::RET_OK;
+}
+
+mindspore::session::LiteSession *Compile(mindspore::lite::Model *model) {
+  // Create and init context.
+  auto context = std::make_shared<mindspore::lite::Context>();
+  if (context == nullptr) {
+    std::cerr << "New context failed while." << std::endl;
+    return nullptr;
+  }
+
+  // Create the session.
+  mindspore::session::LiteSession *session = mindspore::session::LiteSession::CreateSession(context.get());
+  if (session == nullptr) {
+    std::cerr << "CreateSession failed while running." << std::endl;
+    return nullptr;
+  }
+
+  // Compile graph.
+  auto ret = session->CompileGraph(model);
+  if (ret != mindspore::lite::RET_OK) {
+    std::cerr << "Compile failed while running." << std::endl;
+    return nullptr;
+  }
+
+  // Note: when use model->Free(), the model can not be compiled again.
+  if (model != nullptr) {
+    model->Free();
+  }
+  return session;
+}
+
+int CompileAndRun(int argc, const char **argv) {
+  if (argc < 2) {
+    std::cerr << "Usage: ./mindspore_quick_start_cpp ../model/mobilenetv2.ms\n";
+    return -1;
+  }
+  // Read model file.
+  auto model_path = argv[1];
+  size_t size = 0;
+  char *model_buf = ReadFile(model_path, &size);
+  if (model_buf == nullptr) {
+    std::cerr << "Read model file failed." << std::endl;
+    return -1;
+  }
+  // Load the .ms model.
+  auto model = mindspore::lite::Model::Import(model_buf, size);
+  delete[](model_buf);
+  if (model == nullptr) {
+    std::cerr << "Import model file failed." << std::endl;
+    return -1;
+  }
+  // Compile MindSpore Lite model.
+  auto session = Compile(model);
+  if (session == nullptr) {
+    std::cerr << "Create session failed." << std::endl;
+    return -1;
+  }
+  // Run inference.
+  auto ret = Run(session);
+  if (ret != mindspore::lite::RET_OK) {
+    std::cerr << "MindSpore Lite run failed." << std::endl;
+    return -1;
+  }
+  // Delete model buffer.
+  delete model;
+  // Delete session buffer.
+  delete session;
+  return mindspore::lite::RET_OK;
+}
+
+int main(int argc, const char **argv) { return CompileAndRun(argc, argv); }
diff --git a/mindspore/lite/examples/runtime_cpp/CMakeLists.txt b/mindspore/lite/examples/runtime_cpp/CMakeLists.txt
new file mode 100644
index 0000000000..b406b9f08d
--- /dev/null
+++ b/mindspore/lite/examples/runtime_cpp/CMakeLists.txt
@@ -0,0 +1,48 @@
+cmake_minimum_required(VERSION 3.14)
+project(RuntimeCpp)
+
+if(CMAKE_CXX_COMPILER_ID STREQUAL "GNU" AND CMAKE_CXX_COMPILER_VERSION VERSION_LESS 7.3.0)
+    message(FATAL_ERROR "GCC version ${CMAKE_CXX_COMPILER_VERSION} must not be less than 7.3.0")
+endif()
+link_directories(${CMAKE_CURRENT_SOURCE_DIR}/lib)
+set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++17")
+
+include_directories(${CMAKE_CURRENT_SOURCE_DIR})
+file(GLOB_RECURSE RUNTIME_CPP ${CMAKE_CURRENT_SOURCE_DIR}/*.cc)
+
+add_executable(runtime_cpp ${RUNTIME_CPP})
+find_library(log-lib log)
+target_link_libraries(
+        runtime_cpp
+        -Wl,--whole-archive mindspore-lite -Wl,--no-whole-archive
+        hiai
+        hiai_ir
+        hiai_ir_build
+        ${log-lib}
+)
+
+SET(CMAKE_INSTALL_PREFIX ${CMAKE_CURRENT_SOURCE_DIR}/build/tmp)
+
+INSTALL(TARGETS runtime_cpp
+        DESTINATION exe)
+
+INSTALL(FILES ${CMAKE_CURRENT_SOURCE_DIR}/lib/libhiai.so
+        DESTINATION lib)
+INSTALL(FILES ${CMAKE_CURRENT_SOURCE_DIR}/lib/libhiai_ir.so
+        DESTINATION lib)
+INSTALL(FILES ${CMAKE_CURRENT_SOURCE_DIR}/lib/libhiai_ir_build.so
+        DESTINATION lib)
+INSTALL(FILES
+        ${ANDROID_NDK}/toolchains/llvm/prebuilt/linux-x86_64/sysroot/usr/lib/aarch64-linux-android/libc++_shared.so
+        DESTINATION lib)
+
+INSTALL(FILES ${CMAKE_CURRENT_SOURCE_DIR}/model/mobilenetv2.ms
+        DESTINATION model)
+
+set(CPACK_GENERATOR "TGZ")
+
+set(CPACK_PACKAGE_FILE_NAME "runtime_cpp_demo")
+
+set(CPACK_PACKAGE_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR}/output)
+
+include(CPack)
diff --git a/mindspore/lite/examples/runtime_cpp/README.md b/mindspore/lite/examples/runtime_cpp/README.md
new file mode 100644
index 0000000000..5b244e68aa
--- /dev/null
+++ b/mindspore/lite/examples/runtime_cpp/README.md
@@ -0,0 +1,55 @@
+## 构建与运行
+
+- 环境要求
+    - 系统环境：Linux x86_64，推荐使用Ubuntu 18.04.02LTS
+    - 编译依赖：
+        - [CMake](https://cmake.org/download/) >= 3.18.3
+        - [GCC](https://gcc.gnu.org/releases.html) >= 7.3.0
+        - [Android_NDK](https://dl.google.com/android/repository/android-ndk-r20b-linux-x86_64.zip) >= r20
+    - [Git](https://git-scm.com/downloads) >= 2.28.0
+    
+- 编译构建
+
+  在`mindspore/lite/examples/runtime_cpp`目录下执行build脚本，将能够自动下载相关文件并编译Demo。
+
+  ```bash
+  bash build.sh
+  ```
+
+  > 若MindSpore Lite推理框架下载失败，请手动下载硬件平台为CPU，操作系统为Ubuntu-x64的[MindSpore Lite 模型推理框架](https://www.mindspore.cn/tutorial/lite/zh-CN/r1.1/use/downloads.html)，解压后将其拷贝对应到`mindspore/lite/examples/runtime_cpp/lib`目录。
+  >
+  > 若mobilenetv2模型下载失败，请手动下载相关模型文件[mobilenetv2](https://download.mindspore.cn/model_zoo/official/lite/mobilenetv2_openimage_lite/mobilenetv2.ms)，并将其拷贝到`mindspore/lite/examples/runtime_cpp/model`目录。
+
+- 文件传输
+
+  使用`adb`将`mindspore/lite/examples/runtime_cpp\output`目录下的`runtime_cpp_demo.tar.gz`压缩包发送到Android手机
+
+  ```shell
+  adb push runtime_cpp_demo.tar.gz /data/local/tmp
+  ```
+
+- 执行推理
+
+  使用`adb`进入Android Shell命令模式
+
+  ```shell
+  adb shell
+  ```
+
+  进入压缩包所在的相关目录，并进行解压
+
+  ```shell
+  cd /data/local/tmp && tar xzvf runtime_cpp_demo.tar.gz
+  ```
+
+  配置`LD_LIBRARY_PATH`环境变量
+
+  ```shell
+  export LD_LIBRARY_PATH = /data/local/tmp/runtime_cpp_demo/lib:{LD_LIBRARY_PATH}
+  ```
+
+  编译构建后，进入`mindspore/lite/examples/runtime_cpp/build`目录，并执行以下命令，体验MindSpore Lite推理mobilenetv2模型。
+
+  ```bash
+  ./runtime_cpp ../model/mobilenetv2.ms 0
+  ```
diff --git a/mindspore/lite/examples/runtime_cpp/build.sh b/mindspore/lite/examples/runtime_cpp/build.sh
new file mode 100644
index 0000000000..0bafae0890
--- /dev/null
+++ b/mindspore/lite/examples/runtime_cpp/build.sh
@@ -0,0 +1,49 @@
+#!/bin/bash
+# Copyright 2021 Huawei Technologies Co., Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ============================================================================
+
+BASEPATH=$(
+  cd "$(dirname $0)"
+  pwd
+)
+get_version() {
+  VERSION_MAJOR=$(grep "const int ms_version_major =" ${BASEPATH}/../../include/version.h | tr -dc "[0-9]")
+  VERSION_MINOR=$(grep "const int ms_version_minor =" ${BASEPATH}/../../include/version.h | tr -dc "[0-9]")
+  VERSION_REVISION=$(grep "const int ms_version_revision =" ${BASEPATH}/../../include/version.h | tr -dc "[0-9]")
+  VERSION_STR=${VERSION_MAJOR}.${VERSION_MINOR}.${VERSION_REVISION}
+}
+get_version
+MODEL_DOWNLOAD_URL="https://download.mindspore.cn/model_zoo/official/lite/mobilenetv2_openimage_lite/1.1/mobilenetv2.ms"
+MINDSPORE_LITE_DOWNLOAD_URL="https://ms-release.obs.cn-north-4.myhuaweicloud.com/${VERSION_STR}/MindSpore/lite/release/android/mindspore-lite-${VERSION_STR}-inference-android.tar.gz"
+
+mkdir -p build
+mkdir -p lib
+mkdir -p model
+if [ ! -e ${BASEPATH}/model/mobilenetv2.ms ]; then
+  wget -c -O ${BASEPATH}/model/mobilenetv2.ms --no-check-certificate ${MODEL_DOWNLOAD_URL}
+fi
+if [ ! -e ${BASEPATH}/build/mindspore-lite-${VERSION_STR}-inference-android.tar.gz ]; then
+  wget -c -O ${BASEPATH}/build/mindspore-lite-${VERSION_STR}-inference-android.tar.gz --no-check-certificate ${MINDSPORE_LITE_DOWNLOAD_URL}
+fi
+tar xzvf ${BASEPATH}/build/mindspore-lite-${VERSION_STR}-inference-android.tar.gz -C ${BASEPATH}/build/
+cp -r ${BASEPATH}/build/mindspore-lite-${VERSION_STR}-inference-android/lib/aarch64/libmindspore-lite.a ${BASEPATH}/lib
+cp -r ${BASEPATH}/build/mindspore-lite-${VERSION_STR}-inference-android/third_party/hiai_ddk/lib/aarch64/*.so ${BASEPATH}/lib
+cp -r ${BASEPATH}/build/mindspore-lite-${VERSION_STR}-inference-android/include ${BASEPATH}/
+
+cd ${BASEPATH}/build
+cmake -DCMAKE_TOOLCHAIN_FILE="${ANDROID_NDK}/build/cmake/android.toolchain.cmake" -DANDROID_NATIVE_API_LEVEL="19" \
+  -DANDROID_NDK="${ANDROID_NDK}" -DANDROID_ABI="arm64-v8a" -DANDROID_STL="c++_shared" ${BASEPATH}
+
+make && make install && make package
diff --git a/mindspore/lite/examples/runtime_cpp/main.cc b/mindspore/lite/examples/runtime_cpp/main.cc
new file mode 100644
index 0000000000..08c005e3da
--- /dev/null
+++ b/mindspore/lite/examples/runtime_cpp/main.cc
@@ -0,0 +1,699 @@
+/**
+ * Copyright 2021 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include <iostream>
+#include <cstring>
+#include <random>
+#include <fstream>
+#include <thread>
+#include <algorithm>
+#include "include/errorcode.h"
+#include "include/model.h"
+#include "include/context.h"
+#include "include/lite_session.h"
+#include "include/version.h"
+
+char *ReadFile(const char *file, size_t *size) {
+  if (file == nullptr) {
+    std::cerr << "file is nullptr." << std::endl;
+    return nullptr;
+  }
+
+  std::ifstream ifs(file);
+  if (!ifs.good()) {
+    std::cerr << "file: " << file << " is not exist." << std::endl;
+    return nullptr;
+  }
+
+  if (!ifs.is_open()) {
+    std::cerr << "file: " << file << " open failed." << std::endl;
+    return nullptr;
+  }
+
+  ifs.seekg(0, std::ios::end);
+  *size = ifs.tellg();
+  std::unique_ptr<char[]> buf(new (std::nothrow) char[*size]);
+  if (buf == nullptr) {
+    std::cerr << "malloc buf failed, file: " << file << std::endl;
+    ifs.close();
+    return nullptr;
+  }
+
+  ifs.seekg(0, std::ios::beg);
+  ifs.read(buf.get(), *size);
+  ifs.close();
+
+  return buf.release();
+}
+
+template <typename T, typename Distribution>
+void GenerateRandomData(int size, void *data, Distribution distribution) {
+  if (data == nullptr) {
+    std::cerr << "data is nullptr." << std::endl;
+    return;
+  }
+  std::mt19937 random_engine;
+  int elements_num = size / sizeof(T);
+  (void)std::generate_n(static_cast<T *>(data), elements_num,
+                        [&]() { return static_cast<T>(distribution(random_engine)); });
+}
+
+std::shared_ptr<mindspore::lite::Context> CreateCPUContext() {
+  auto context = std::make_shared<mindspore::lite::Context>();
+  if (context == nullptr) {
+    std::cerr << "New context failed while running." << std::endl;
+    return nullptr;
+  }
+  // CPU device context has default values.
+  auto &cpu_device_info = context->device_list_[0].device_info_.cpu_device_info_;
+  // The large core takes priority in thread and core binding methods. This parameter will work in the BindThread
+  // interface. For specific binding effect, see the "Run Graph" section.
+  cpu_device_info.cpu_bind_mode_ = mindspore::lite::HIGHER_CPU;
+  // Use float16 operator as priority.
+  cpu_device_info.enable_float16_ = true;
+  return context;
+}
+
+std::shared_ptr<mindspore::lite::Context> CreateGPUContext() {
+  auto context = std::make_shared<mindspore::lite::Context>();
+  if (context == nullptr) {
+    std::cerr << "New context failed while running. " << std::endl;
+    return nullptr;
+  }
+
+  // If GPU device context is set. The preferred backend is GPU, which means, if there is a GPU operator, it will run on
+  // the GPU first, otherwise it will run on the CPU.
+  mindspore::lite::DeviceContext gpu_device_ctx{mindspore::lite::DT_GPU, {false}};
+  // GPU use float16 operator as priority.
+  gpu_device_ctx.device_info_.gpu_device_info_.enable_float16_ = true;
+  // The GPU device context needs to be push_back into device_list to work.
+  context->device_list_.push_back(gpu_device_ctx);
+  return context;
+}
+
+std::shared_ptr<mindspore::lite::Context> CreateNPUContext() {
+  auto context = std::make_shared<mindspore::lite::Context>();
+  if (context == nullptr) {
+    std::cerr << "New context failed while running. " << std::endl;
+    return nullptr;
+  }
+  mindspore::lite::DeviceContext npu_device_ctx{mindspore::lite::DT_NPU};
+  npu_device_ctx.device_info_.npu_device_info_.frequency_ = 3;
+  // The NPU device context needs to be push_back into device_list to work.
+  context->device_list_.push_back(npu_device_ctx);
+  return context;
+}
+
+int GetInputsAndSetData(mindspore::session::LiteSession *session) {
+  auto inputs = session->GetInputs();
+
+  // The model has only one input tensor.
+  auto in_tensor = inputs.front();
+  if (in_tensor == nullptr) {
+    std::cerr << "Input tensor is nullptr" << std::endl;
+    return -1;
+  }
+  auto input_data = in_tensor->MutableData();
+  if (input_data == nullptr) {
+    std::cerr << "MallocData for inTensor failed." << std::endl;
+    return -1;
+  }
+  GenerateRandomData<float>(in_tensor->Size(), input_data, std::uniform_real_distribution<float>(0.1f, 1.0f));
+
+  return 0;
+}
+
+int GetInputsByTensorNameAndSetData(mindspore::session::LiteSession *session) {
+  auto in_tensor = session->GetInputsByTensorName("2029_2028_1_construct_wrapper:x");
+  if (in_tensor == nullptr) {
+    std::cerr << "Input tensor is nullptr" << std::endl;
+    return -1;
+  }
+  auto input_data = in_tensor->MutableData();
+  if (input_data == nullptr) {
+    std::cerr << "MallocData for inTensor failed." << std::endl;
+    return -1;
+  }
+  GenerateRandomData<float>(in_tensor->Size(), input_data, std::uniform_real_distribution<float>(0.1f, 1.0f));
+  return 0;
+}
+
+void GetOutputsByNodeName(mindspore::session::LiteSession *session) {
+  // model has a output node named output_node_name_0.
+  auto output_vec = session->GetOutputsByNodeName("Default/Sigmoid-op204");
+  // output node named output_node_name_0 has only one output tensor.
+  auto out_tensor = output_vec.front();
+  if (out_tensor == nullptr) {
+    std::cerr << "Output tensor is nullptr" << std::endl;
+    return;
+  }
+  std::cout << "tensor size is:" << out_tensor->Size() << " tensor elements num is:" << out_tensor->ElementsNum()
+            << std::endl;
+  // The model output data is float 32.
+  if (out_tensor->data_type() != mindspore::TypeId::kNumberTypeFloat32) {
+    std::cerr << "Output should in float32" << std::endl;
+    return;
+  }
+  auto out_data = reinterpret_cast<float *>(out_tensor->MutableData());
+  if (out_data == nullptr) {
+    std::cerr << "Data of out_tensor is nullptr" << std::endl;
+    return;
+  }
+  std::cout << "output data is:";
+  for (int i = 0; i < out_tensor->ElementsNum() && i < 10; i++) {
+    std::cout << out_data[i] << " ";
+  }
+  std::cout << std::endl;
+}
+
+void GetOutputByTensorName(mindspore::session::LiteSession *session) {
+  // We can use GetOutputTensorNames method to get all name of output tensor of model which is in order.
+  auto tensor_names = session->GetOutputTensorNames();
+  // Use output tensor name returned by GetOutputTensorNames as key
+  for (const auto &tensor_name : tensor_names) {
+    auto out_tensor = session->GetOutputByTensorName(tensor_name);
+    if (out_tensor == nullptr) {
+      std::cerr << "Output tensor is nullptr" << std::endl;
+      return;
+    }
+    std::cout << "tensor size is:" << out_tensor->Size() << " tensor elements num is:" << out_tensor->ElementsNum()
+              << std::endl;
+    // The model output data is float 32.
+    if (out_tensor->data_type() != mindspore::TypeId::kNumberTypeFloat32) {
+      std::cerr << "Output should in float32" << std::endl;
+      return;
+    }
+    auto out_data = reinterpret_cast<float *>(out_tensor->MutableData());
+    if (out_data == nullptr) {
+      std::cerr << "Data of out_tensor is nullptr" << std::endl;
+      return;
+    }
+    std::cout << "output data is:";
+    for (int i = 0; i < out_tensor->ElementsNum() && i < 10; i++) {
+      std::cout << out_data[i] << " ";
+    }
+    std::cout << std::endl;
+  }
+}
+
+void GetOutputs(mindspore::session::LiteSession *session) {
+  auto out_tensors = session->GetOutputs();
+  for (auto out_tensor : out_tensors) {
+    std::cout << "tensor name is:" << out_tensor.first << " tensor size is:" << out_tensor.second->Size()
+              << " tensor elements num is:" << out_tensor.second->ElementsNum() << std::endl;
+    // The model output data is float 32.
+    if (out_tensor.second->data_type() != mindspore::TypeId::kNumberTypeFloat32) {
+      std::cerr << "Output should in float32" << std::endl;
+      return;
+    }
+    auto out_data = reinterpret_cast<float *>(out_tensor.second->MutableData());
+    if (out_data == nullptr) {
+      std::cerr << "Data of out_tensor is nullptr" << std::endl;
+      return;
+    }
+    std::cout << "output data is:";
+    for (int i = 0; i < out_tensor.second->ElementsNum() && i < 10; i++) {
+      std::cout << out_data[i] << " ";
+    }
+    std::cout << std::endl;
+  }
+}
+
+mindspore::session::LiteSession *CreateSessionAndCompileByModel(mindspore::lite::Model *model) {
+  // Create and init CPU context.
+  // If you need to use GPU or NPU, you can refer to CreateGPUContext() or CreateNPUContext().
+  auto context = CreateCPUContext();
+  if (context == nullptr) {
+    std::cerr << "New context failed while." << std::endl;
+    return nullptr;
+  }
+
+  // Create the session.
+  mindspore::session::LiteSession *session = mindspore::session::LiteSession::CreateSession(context.get());
+  if (session == nullptr) {
+    std::cerr << "CreateSession failed while running." << std::endl;
+    return nullptr;
+  }
+
+  // Compile graph.
+  auto ret = session->CompileGraph(model);
+  if (ret != mindspore::lite::RET_OK) {
+    delete session;
+    std::cerr << "Compile failed while running." << std::endl;
+    return nullptr;
+  }
+
+  return session;
+}
+
+mindspore::session::LiteSession *CreateSessionAndCompileByModelBuffer(char *model_buf, size_t size) {
+  auto context = std::make_shared<mindspore::lite::Context>();
+  if (context == nullptr) {
+    std::cerr << "New context failed while running" << std::endl;
+    return nullptr;
+  }
+  // Use model buffer and context to create Session.
+  auto session = mindspore::session::LiteSession::CreateSession(model_buf, size, context.get());
+  if (session == nullptr) {
+    std::cerr << "CreateSession failed while running" << std::endl;
+    return nullptr;
+  }
+  return session;
+}
+
+int ResizeInputsTensorShape(mindspore::session::LiteSession *session) {
+  auto inputs = session->GetInputs();
+  std::vector<int> resize_shape = {1, 128, 128, 3};
+  // Assume the model has only one input,resize input shape to [1, 128, 128, 3]
+  std::vector<std::vector<int>> new_shapes;
+  new_shapes.push_back(resize_shape);
+  return session->Resize(inputs, new_shapes);
+}
+
+int Run(const char *model_path) {
+  // Read model file.
+  size_t size = 0;
+  char *model_buf = ReadFile(model_path, &size);
+  if (model_buf == nullptr) {
+    std::cerr << "Read model file failed." << std::endl;
+    return -1;
+  }
+  // Load the .ms model.
+  auto model = mindspore::lite::Model::Import(model_buf, size);
+  delete[](model_buf);
+  if (model == nullptr) {
+    std::cerr << "Import model file failed." << std::endl;
+    return -1;
+  }
+  // Compile MindSpore Lite model.
+  auto session = CreateSessionAndCompileByModel(model);
+  if (session == nullptr) {
+    delete model;
+    std::cerr << "Create session failed." << std::endl;
+    return -1;
+  }
+
+  // Note: when use model->Free(), the model can not be compiled again.
+  model->Free();
+
+  // Set inputs data.
+  // You can also get input through other methods, and you can refer to GetInputsAndSetData()
+  GetInputsByTensorNameAndSetData(session);
+
+  session->BindThread(true);
+  auto ret = session->RunGraph();
+  if (ret != mindspore::lite::RET_OK) {
+    delete model;
+    std::cerr << "Inference error " << ret << std::endl;
+    return ret;
+  }
+  session->BindThread(false);
+
+  // Get outputs data.
+  // You can also get output through other methods,
+  // and you can refer to GetOutputByTensorName() or GetOutputs().
+  GetOutputsByNodeName(session);
+
+  // Delete model buffer.
+  delete model;
+  // Delete session buffer.
+  delete session;
+  return 0;
+}
+
+int RunResize(const char *model_path) {
+  size_t size = 0;
+  char *model_buf = ReadFile(model_path, &size);
+  if (model_buf == nullptr) {
+    std::cerr << "Read model file failed." << std::endl;
+    return -1;
+  }
+  // Load the .ms model.
+  auto model = mindspore::lite::Model::Import(model_buf, size);
+  delete[](model_buf);
+  if (model == nullptr) {
+    std::cerr << "Import model file failed." << std::endl;
+    return -1;
+  }
+  // Compile MindSpore Lite model.
+  auto session = CreateSessionAndCompileByModel(model);
+  if (session == nullptr) {
+    delete model;
+    std::cerr << "Create session failed." << std::endl;
+    return -1;
+  }
+
+  // Resize inputs tensor shape.
+  auto ret = ResizeInputsTensorShape(session);
+  if (ret != mindspore::lite::RET_OK) {
+    delete model;
+    std::cerr << "Resize input tensor shape error." << ret << std::endl;
+    return ret;
+  }
+
+  // Set inputs data.
+  // You can also get input through other methods, and you can refer to GetInputsAndSetData()
+  GetInputsByTensorNameAndSetData(session);
+
+  session->BindThread(true);
+  ret = session->RunGraph();
+  if (ret != mindspore::lite::RET_OK) {
+    delete model;
+    std::cerr << "Inference error " << ret << std::endl;
+    return ret;
+  }
+  session->BindThread(false);
+
+  // Get outputs data.
+  // You can also get output through other methods,
+  // and you can refer to GetOutputByTensorName() or GetOutputs().
+  GetOutputsByNodeName(session);
+
+  // Delete model buffer.
+  delete model;
+  // Delete session buffer.
+  delete session;
+  return 0;
+}
+
+int RunCreateSessionSimplified(const char *model_path) {
+  size_t size = 0;
+  char *model_buf = ReadFile(model_path, &size);
+  if (model_buf == nullptr) {
+    std::cerr << "Read model file failed." << std::endl;
+    return -1;
+  }
+
+  // Compile MindSpore Lite model.
+  auto session = CreateSessionAndCompileByModelBuffer(model_buf, size);
+  if (session == nullptr) {
+    std::cerr << "Create session failed." << std::endl;
+    return -1;
+  }
+
+  // Set inputs data.
+  // You can also get input through other methods, and you can refer to GetInputsAndSetData()
+  GetInputsByTensorNameAndSetData(session);
+
+  session->BindThread(true);
+  auto ret = session->RunGraph();
+  if (ret != mindspore::lite::RET_OK) {
+    std::cerr << "Inference error " << ret << std::endl;
+    return ret;
+  }
+  session->BindThread(false);
+
+  // Get outputs data.
+  // You can also get output through other methods,
+  // and you can refer to GetOutputByTensorName() or GetOutputs().
+  GetOutputsByNodeName(session);
+
+  // Delete session buffer.
+  delete session;
+  return 0;
+}
+
+int RunSessionParallel(const char *model_path) {
+  size_t size = 0;
+  char *model_buf = ReadFile(model_path, &size);
+  if (model_buf == nullptr) {
+    std::cerr << "Read model file failed." << std::endl;
+    return -1;
+  }
+  // Load the .ms model.
+  auto model = mindspore::lite::Model::Import(model_buf, size);
+  delete[](model_buf);
+  if (model == nullptr) {
+    std::cerr << "Import model file failed." << std::endl;
+    return -1;
+  }
+  // Compile MindSpore Lite model.
+  auto session1 = CreateSessionAndCompileByModel(model);
+  if (session1 == nullptr) {
+    delete model;
+    std::cerr << "Create session failed." << std::endl;
+    return -1;
+  }
+
+  // Compile MindSpore Lite model.
+  auto session2 = CreateSessionAndCompileByModel(model);
+  if (session2 == nullptr) {
+    delete model;
+    std::cerr << "Create session failed." << std::endl;
+    return -1;
+  }
+  // Note: when use model->Free(), the model can not be compiled again.
+  model->Free();
+
+  std::thread thread1([&]() {
+    GetInputsByTensorNameAndSetData(session1);
+    auto status = session1->RunGraph();
+    if (status != 0) {
+      if (model != nullptr) {
+        delete model;
+        model = nullptr;
+      }
+      std::cerr << "Inference error " << status << std::endl;
+      return;
+    }
+    std::cout << "Session1 inference success" << std::endl;
+  });
+
+  std::thread thread2([&]() {
+    GetInputsByTensorNameAndSetData(session2);
+    auto status = session2->RunGraph();
+    if (status != 0) {
+      if (model != nullptr) {
+        delete model;
+        model = nullptr;
+      }
+      std::cerr << "Inference error " << status << std::endl;
+      return;
+    }
+    std::cout << "Session2 inference success" << std::endl;
+  });
+
+  thread1.join();
+  thread2.join();
+
+  // Get outputs data.
+  // You can also get output through other methods,
+  // and you can refer to GetOutputByTensorName() or GetOutputs().
+  GetOutputsByNodeName(session1);
+  GetOutputsByNodeName(session2);
+
+  // Delete model buffer.
+  if (model != nullptr) {
+    delete model;
+    model = nullptr;
+  }
+  // Delete session buffer.
+  delete session1;
+  delete session2;
+  return 0;
+}
+
+int RunWithSharedMemoryPool(const char *model_path) {
+  size_t size = 0;
+  char *model_buf = ReadFile(model_path, &size);
+  if (model_buf == nullptr) {
+    std::cerr << "Read model file failed." << std::endl;
+    return -1;
+  }
+  auto model = mindspore::lite::Model::Import(model_buf, size);
+  delete[](model_buf);
+  if (model == nullptr) {
+    std::cerr << "Import model file failed." << std::endl;
+    return -1;
+  }
+
+  auto context1 = std::make_shared<mindspore::lite::Context>();
+  if (context1 == nullptr) {
+    delete model;
+    std::cerr << "New context failed while running." << std::endl;
+    return -1;
+  }
+  auto session1 = mindspore::session::LiteSession::CreateSession(context1.get());
+
+  if (session1 == nullptr) {
+    delete model;
+    std::cerr << "CreateSession failed while running." << std::endl;
+    return -1;
+  }
+  auto ret = session1->CompileGraph(model);
+  if (ret != mindspore::lite::RET_OK) {
+    delete model;
+    std::cerr << "Compile failed while running." << std::endl;
+    return -1;
+  }
+
+  auto context2 = std::make_shared<mindspore::lite::Context>();
+  if (context2 == nullptr) {
+    delete model;
+    std::cerr << "New  context failed while running." << std::endl;
+    return -1;
+  }
+  // Use the same allocator to share the memory pool.
+  context2->allocator = context1->allocator;
+
+  auto session2 = mindspore::session::LiteSession::CreateSession(context2.get());
+  if (session2 == nullptr) {
+    delete model;
+    std::cerr << "CreateSession failed while running " << std::endl;
+    return -1;
+  }
+
+  ret = session2->CompileGraph(model);
+  if (ret != mindspore::lite::RET_OK) {
+    delete model;
+    std::cerr << "Compile failed while running " << std::endl;
+    return -1;
+  }
+
+  // Note: when use model->Free(), the model can not be compiled again.
+  model->Free();
+
+  // Set inputs data.
+  // You can also get input through other methods, and you can refer to GetInputsAndSetData()
+  GetInputsByTensorNameAndSetData(session1);
+  GetInputsByTensorNameAndSetData(session2);
+
+  ret = session1->RunGraph();
+  if (ret != mindspore::lite::RET_OK) {
+    std::cerr << "Inference error " << ret << std::endl;
+    return ret;
+  }
+
+  ret = session2->RunGraph();
+  if (ret != mindspore::lite::RET_OK) {
+    delete model;
+    std::cerr << "Inference error " << ret << std::endl;
+    return ret;
+  }
+
+  // Get outputs data.
+  // You can also get output through other methods,
+  // and you can refer to GetOutputByTensorName() or GetOutputs().
+  GetOutputsByNodeName(session1);
+  GetOutputsByNodeName(session2);
+
+  // Delete model buffer.
+  delete model;
+  // Delete session buffer.
+  delete session1;
+  delete session2;
+  return 0;
+}
+
+int RunCallback(const char *model_path) {
+  size_t size = 0;
+  char *model_buf = ReadFile(model_path, &size);
+  if (model_buf == nullptr) {
+    std::cerr << "Read model file failed." << std::endl;
+    return -1;
+  }
+  // Load the .ms model.
+  auto model = mindspore::lite::Model::Import(model_buf, size);
+  delete[](model_buf);
+  if (model == nullptr) {
+    std::cerr << "Import model file failed." << std::endl;
+    return -1;
+  }
+  // Compile MindSpore Lite model.
+  auto session = CreateSessionAndCompileByModel(model);
+  if (session == nullptr) {
+    delete model;
+    std::cerr << "Create session failed." << std::endl;
+    return -1;
+  }
+
+  // Note: when use model->Free(), the model can not be compiled again.
+  model->Free();
+
+  // Set inputs data.
+  // You can also get input through other methods, and you can refer to GetInputsAndSetData()
+  GetInputsByTensorNameAndSetData(session);
+
+  // Definition of callback function before forwarding operator.
+  auto before_call_back = [&](const std::vector<mindspore::tensor::MSTensor *> &before_inputs,
+                              const std::vector<mindspore::tensor::MSTensor *> &before_outputs,
+                              const mindspore::CallBackParam &call_param) {
+    std::cout << "Before forwarding " << call_param.node_name << " " << call_param.node_type << std::endl;
+    return true;
+  };
+  // Definition of callback function after forwarding operator.
+  auto after_call_back = [&](const std::vector<mindspore::tensor::MSTensor *> &after_inputs,
+                             const std::vector<mindspore::tensor::MSTensor *> &after_outputs,
+                             const mindspore::CallBackParam &call_param) {
+    std::cout << "After forwarding " << call_param.node_name << " " << call_param.node_type << std::endl;
+    return true;
+  };
+
+  session->BindThread(true);
+  auto ret = session->RunGraph(before_call_back, after_call_back);
+  if (ret != mindspore::lite::RET_OK) {
+    delete model;
+    std::cerr << "Inference error " << ret << std::endl;
+    return ret;
+  }
+  session->BindThread(false);
+
+  // Get outputs data.
+  // You can also get output through other methods,
+  // and you can refer to GetOutputByTensorName() or GetOutputs().
+  GetOutputsByNodeName(session);
+
+  // Delete model buffer.
+  delete model;
+  // Delete session buffer.
+  delete session;
+  return 0;
+}
+
+int main(int argc, const char **argv) {
+  if (argc < 3) {
+    std::cerr << "Usage: ./runtime_cpp model_path flag" << std::endl;
+    std::cerr << "Example: ./runtime_cpp ../model/mobilenetv2.ms 0" << std::endl;
+    std::cerr << "When your Flag is 0, you will run MindSpore Lite inference." << std::endl;
+    std::cerr << "When your Flag is 1, you will run MindSpore Lite inference with resize." << std::endl;
+    std::cerr << "When your Flag is 2, you will run MindSpore Lite inference with CreateSession simplified API."
+              << std::endl;
+    std::cerr << "When your Flag is 3, you will run MindSpore Lite inference with session parallel." << std::endl;
+    std::cerr << "When your Flag is 4, you will run MindSpore Lite inference with shared memory pool." << std::endl;
+    std::cerr << "When your Flag is 5, you will run MindSpore Lite inference with callback." << std::endl;
+    return -1;
+  }
+  std::string version = mindspore::lite::Version();
+  std::cout << "MindSpore Lite Version is " << version << std::endl;
+  auto model_path = argv[1];
+  auto flag = argv[2];
+  if (strcmp(flag, "0") == 0) {
+    return Run(model_path);
+  } else if (strcmp(flag, "1") == 0) {
+    return RunResize(model_path);
+  } else if (strcmp(flag, "2") == 0) {
+    return RunCreateSessionSimplified(model_path);
+  } else if (strcmp(flag, "3") == 0) {
+    return RunSessionParallel(model_path);
+  } else if (strcmp(flag, "4") == 0) {
+    return RunWithSharedMemoryPool(model_path);
+  } else if (strcmp(flag, "5") == 0) {
+    return RunCallback(model_path);
+  } else {
+    std::cerr << "Unsupported Flag " << flag << std::endl;
+    return -1;
+  }
+}