ncf && textrcnn && bgcf && tinybert 310 infer

modified: official/recommend/ncf/preprocess.py
2021-06-01 17:35:27 +08:00 · 2021-06-01 17:35:27 +08:00 · 3df895f477
parent ad165deb15
commit 3df895f477
39 changed files with 2645 additions and 4 deletions
--- a/model_zoo/official/gnn/bgcf/README.md
+++ b/model_zoo/official/gnn/bgcf/README.md
@ -16,6 +16,10 @@
        - [Training](#training)
    - [Evaluation Process](#evaluation-process)
        - [Evaluation](#evaluation)
+    - [Inference Process](#inference-process)
+        - [Export MindIR](#export-mindir)
+        - [Infer on Ascend310](#infer-on-ascend310)
+        - [result](#result)
 - [Model Description](#model-description)
    - [Performance](#performance)
 - [Description of random situation](#description-of-random-situation)
@ -244,6 +248,38 @@ Parameters for both training and evaluation can be set in config.py.
  sedp_@10:0.01926,     sedp_@20:0.01547,    nov_@10:7.60851,    nov_@20:7.81969
  ```

+## Inference Process
+
+### [Export MindIR](#contents)
+
+```shell
+python export.py --ckpt_file [CKPT_PATH] --file_name [FILE_NAME] --file_format [FILE_FORMAT]
+```
+
+The ckpt_file parameter is required,
+`EXPORT_FORMAT` should be in ["AIR", "MINDIR"]
+
+### Infer on Ascend310
+
+Before performing inference, the mindir file must be exported by `export.py` script. We only provide an example of inference using MINDIR model.
+
+```shell
+# Ascend310 inference
+bash run_infer_310.sh [MINDIR_PATH] [DATASET_NAME] [NEED_PREPROCESS] [DEVICE_ID]
+```
+
+- `NEED_PREPROCESS` means weather need preprocess or not, it's value is 'y' or 'n'.
+- `DEVICE_ID` is optional, default value is 0.
+
+### result
+
+Inference result is saved in current path, you can find result like this in acc.log file.
+
+```bash
+recall_@10:0.10383,     recall_@20:0.15524,     ndcg_@10:0.07503,    ndcg_@20:0.09249,
+  sedp_@10:0.01926,     sedp_@20:0.01547,    nov_@10:7.60851,    nov_@20:7.81969
+```
+
 ## [Model Description](#contents)

 ### [Performance](#contents)
--- a/model_zoo/official/gnn/bgcf/README_CN.md
+++ b/model_zoo/official/gnn/bgcf/README_CN.md
@ -17,6 +17,10 @@
        - [训练](#训练)
    - [评估过程](#评估过程)
        - [评估](#评估)
+    - [推理过程](#推理过程)
+        - [导出MindIR](#导出mindir)
+        - [在Ascend310执行推理](#在ascend310执行推理)
+        - [结果](#结果)
 - [模型描述](#模型描述)
    - [性能](#性能)
 - [随机情况说明](#随机情况说明)
@ -271,6 +275,38 @@ BGCF包含两个主要模块。首先是抽样，它生成基于节点复制的

  ```

+## 推理过程
+
+### [导出MindIR](#contents)
+
+```shell
+python export.py --ckpt_file [CKPT_PATH] --file_name [FILE_NAME] --file_format [FILE_FORMAT]
+```
+
+参数ckpt_file为必填项，
+`EXPORT_FORMAT` 必须在 ["AIR", "MINDIR"]中选择。
+
+### 在Ascend310执行推理
+
+在执行推理前，mindir文件必须通过`export.py`脚本导出。以下展示了使用minir模型执行推理的示例。
+
+```shell
+# Ascend310 inference
+bash run_infer_310.sh [MINDIR_PATH] [DATASET_NAME] [NEED_PREPROCESS] [DEVICE_ID]
+```
+
+- `NEED_PREPROCESS` 表示数据是否需要预处理，取值范围为 'y' 或者 'n'。
+- `DEVICE_ID` 可选，默认值为0。
+
+### 结果
+
+推理结果保存在脚本执行的当前路径，你可以在acc.log中看到以下精度计算结果。
+
+```bash
+recall_@10:0.10383,     recall_@20:0.15524,     ndcg_@10:0.07503,    ndcg_@20:0.09249,
+  sedp_@10:0.01926,     sedp_@20:0.01547,    nov_@10:7.60851,    nov_@20:7.81969
+```
+
 ## 模型描述

 ### 训练性能
--- a/model_zoo/official/gnn/bgcf/ascend310_infer/CMakeLists.txt
+++ b/model_zoo/official/gnn/bgcf/ascend310_infer/CMakeLists.txt
@ -0,0 +1,14 @@
+cmake_minimum_required(VERSION 3.14.1)
+project(Ascend310Infer)
+add_compile_definitions(_GLIBCXX_USE_CXX11_ABI=0)
+set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -O2 -g -std=c++17 -Werror -Wall -fPIE -Wl,--allow-shlib-undefined")
+set(PROJECT_SRC_ROOT ${CMAKE_CURRENT_LIST_DIR}/)
+option(MINDSPORE_PATH "mindspore install path" "")
+include_directories(${MINDSPORE_PATH})
+include_directories(${MINDSPORE_PATH}/include)
+include_directories(${PROJECT_SRC_ROOT})
+find_library(MS_LIB libmindspore.so ${MINDSPORE_PATH}/lib)
+file(GLOB_RECURSE MD_LIB ${MINDSPORE_PATH}/_c_dataengine*)
+
+add_executable(main src/main.cc src/utils.cc)
+target_link_libraries(main ${MS_LIB} ${MD_LIB} gflags)
--- a/model_zoo/official/gnn/bgcf/ascend310_infer/build.sh
+++ b/model_zoo/official/gnn/bgcf/ascend310_infer/build.sh
@ -0,0 +1,29 @@
+#!/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.
+# ============================================================================
+if [ -d out ]; then
+    rm -rf out
+fi
+
+mkdir out
+cd out || exit
+
+if [ -f "Makefile" ]; then
+  make clean
+fi
+
+cmake .. \
+    -DMINDSPORE_PATH="`pip3.7 show mindspore-ascend | grep Location | awk '{print $2"/mindspore"}' | xargs realpath`"
+make
--- a/model_zoo/official/gnn/bgcf/ascend310_infer/inc/utils.h
+++ b/model_zoo/official/gnn/bgcf/ascend310_infer/inc/utils.h
@ -0,0 +1,32 @@
+/**
+ * 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.
+ */
+
+#ifndef MINDSPORE_INFERENCE_UTILS_H_
+#define MINDSPORE_INFERENCE_UTILS_H_
+
+#include <sys/stat.h>
+#include <dirent.h>
+#include <vector>
+#include <string>
+#include <memory>
+#include "include/api/types.h"
+
+std::vector<std::string> GetAllFiles(std::string_view dirName);
+DIR *OpenDir(std::string_view dirName);
+std::string RealPath(std::string_view path);
+mindspore::MSTensor ReadFileToTensor(const std::string &file);
+int WriteResult(const std::string& imageFile, const std::vector<mindspore::MSTensor> &outputs);
+#endif
--- a/model_zoo/official/gnn/bgcf/ascend310_infer/src/main.cc
+++ b/model_zoo/official/gnn/bgcf/ascend310_infer/src/main.cc
@ -0,0 +1,160 @@
+/**
+ * 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 <sys/time.h>
+#include <gflags/gflags.h>
+#include <dirent.h>
+#include <iostream>
+#include <string>
+#include <algorithm>
+#include <iosfwd>
+#include <vector>
+#include <fstream>
+#include <sstream>
+
+#include "include/api/model.h"
+#include "include/api/context.h"
+#include "include/api/types.h"
+#include "include/api/serialization.h"
+#include "include/dataset/execute.h"
+#include "include/dataset/vision.h"
+#include "inc/utils.h"
+
+using mindspore::Context;
+using mindspore::Serialization;
+using mindspore::Model;
+using mindspore::Status;
+using mindspore::MSTensor;
+using mindspore::dataset::Execute;
+using mindspore::ModelType;
+using mindspore::GraphCell;
+using mindspore::kSuccess;
+
+DEFINE_string(mindir_path, "", "mindir path");
+DEFINE_string(input0_path, ".", "input0 path");
+DEFINE_string(input1_path, ".", "input1 path");
+DEFINE_string(input2_path, ".", "input2 path");
+DEFINE_string(input3_path, ".", "input3 path");
+DEFINE_string(input4_path, ".", "input4 path");
+DEFINE_string(input5_path, ".", "input5 path");
+DEFINE_string(input6_path, ".", "input6 path");
+DEFINE_int32(device_id, 0, "device id");
+
+int main(int argc, char **argv) {
+  gflags::ParseCommandLineFlags(&argc, &argv, true);
+  if (RealPath(FLAGS_mindir_path).empty()) {
+    std::cout << "Invalid mindir" << std::endl;
+    return 1;
+  }
+
+  auto context = std::make_shared<Context>();
+  auto ascend310 = std::make_shared<mindspore::Ascend310DeviceInfo>();
+  ascend310->SetDeviceID(FLAGS_device_id);
+  context->MutableDeviceInfo().push_back(ascend310);
+  mindspore::Graph graph;
+  Serialization::Load(FLAGS_mindir_path, ModelType::kMindIR, &graph);
+
+  Model model;
+  Status ret = model.Build(GraphCell(graph), context);
+  if (ret != kSuccess) {
+    std::cout << "ERROR: Build failed." << std::endl;
+    return 1;
+  }
+
+  std::vector<MSTensor> model_inputs = model.GetInputs();
+  if (model_inputs.empty()) {
+    std::cout << "Invalid model, inputs is empty." << std::endl;
+    return 1;
+  }
+
+  auto input0_files = GetAllFiles(FLAGS_input0_path);
+  auto input1_files = GetAllFiles(FLAGS_input1_path);
+  auto input2_files = GetAllFiles(FLAGS_input2_path);
+  auto input3_files = GetAllFiles(FLAGS_input3_path);
+  auto input4_files = GetAllFiles(FLAGS_input4_path);
+  auto input5_files = GetAllFiles(FLAGS_input5_path);
+  auto input6_files = GetAllFiles(FLAGS_input6_path);
+
+  if (input0_files.empty() || input1_files.empty() || input2_files.empty() ||  input3_files.empty()
+      || input4_files.empty() || input5_files.empty() || input6_files.empty()) {
+    std::cout << "ERROR: input data empty." << std::endl;
+    return 1;
+  }
+
+  std::map<double, double> costTime_map;
+  size_t size = input0_files.size();
+
+  for (size_t i = 0; i < size; ++i) {
+    struct timeval start = {0};
+    struct timeval end = {0};
+    double startTimeMs, endTimeMs;
+    std::vector<MSTensor> inputs, outputs;
+    std::cout << "Start predict input files:" << input0_files[i] << std::endl;
+
+    auto input0 = ReadFileToTensor(input0_files[i]);
+    auto input1 = ReadFileToTensor(input1_files[i]);
+    auto input2 = ReadFileToTensor(input2_files[i]);
+    auto input3 = ReadFileToTensor(input3_files[i]);
+    auto input4 = ReadFileToTensor(input4_files[i]);
+    auto input5 = ReadFileToTensor(input5_files[i]);
+    auto input6 = ReadFileToTensor(input6_files[i]);
+
+    inputs.emplace_back(model_inputs[0].Name(), model_inputs[0].DataType(), model_inputs[0].Shape(),
+                        input0.Data().get(), input0.DataSize());
+    inputs.emplace_back(model_inputs[1].Name(), model_inputs[1].DataType(), model_inputs[1].Shape(),
+                        input1.Data().get(), input1.DataSize());
+    inputs.emplace_back(model_inputs[2].Name(), model_inputs[2].DataType(), model_inputs[2].Shape(),
+                        input2.Data().get(), input2.DataSize());
+    inputs.emplace_back(model_inputs[3].Name(), model_inputs[3].DataType(), model_inputs[3].Shape(),
+                        input3.Data().get(), input3.DataSize());
+    inputs.emplace_back(model_inputs[4].Name(), model_inputs[4].DataType(), model_inputs[4].Shape(),
+                        input4.Data().get(), input4.DataSize());
+    inputs.emplace_back(model_inputs[5].Name(), model_inputs[5].DataType(), model_inputs[5].Shape(),
+                        input5.Data().get(), input5.DataSize());
+    inputs.emplace_back(model_inputs[6].Name(), model_inputs[6].DataType(), model_inputs[6].Shape(),
+                        input6.Data().get(), input6.DataSize());
+
+    gettimeofday(&start, nullptr);
+    ret = model.Predict(inputs, &outputs);
+    gettimeofday(&end, nullptr);
+    if (ret != kSuccess) {
+      std::cout << "Predict " << input0_files[i] << " failed." << std::endl;
+      return 1;
+    }
+    startTimeMs = (1.0 * start.tv_sec * 1000000 + start.tv_usec) / 1000;
+    endTimeMs = (1.0 * end.tv_sec * 1000000 + end.tv_usec) / 1000;
+    costTime_map.insert(std::pair<double, double>(startTimeMs, endTimeMs));
+    WriteResult(input0_files[i], outputs);
+  }
+  double average = 0.0;
+  int inferCount = 0;
+
+  for (auto iter = costTime_map.begin(); iter != costTime_map.end(); iter++) {
+    double diff = 0.0;
+    diff = iter->second - iter->first;
+    average += diff;
+    inferCount++;
+  }
+  average = average / inferCount;
+  std::stringstream timeCost;
+  timeCost << "NN inference cost average time: "<< average << " ms of infer_count " << inferCount << std::endl;
+  std::cout << "NN inference cost average time: "<< average << "ms of infer_count " << inferCount << std::endl;
+  std::string fileName = "./time_Result" + std::string("/test_perform_static.txt");
+  std::ofstream fileStream(fileName.c_str(), std::ios::trunc);
+  fileStream << timeCost.str();
+  fileStream.close();
+  costTime_map.clear();
+  return 0;
+}
--- a/model_zoo/official/gnn/bgcf/ascend310_infer/src/utils.cc
+++ b/model_zoo/official/gnn/bgcf/ascend310_infer/src/utils.cc
@ -0,0 +1,129 @@
+/**
+ * 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 <fstream>
+#include <algorithm>
+#include <iostream>
+#include "inc/utils.h"
+
+using mindspore::MSTensor;
+using mindspore::DataType;
+
+std::vector<std::string> GetAllFiles(std::string_view dirName) {
+  struct dirent *filename;
+  DIR *dir = OpenDir(dirName);
+  if (dir == nullptr) {
+    return {};
+  }
+  std::vector<std::string> res;
+  while ((filename = readdir(dir)) != nullptr) {
+    std::string dName = std::string(filename->d_name);
+    if (dName == "." || dName == ".." || filename->d_type != DT_REG) {
+      continue;
+    }
+    res.emplace_back(std::string(dirName) + "/" + filename->d_name);
+  }
+  std::sort(res.begin(), res.end());
+  for (auto &f : res) {
+    std::cout << "image file: " << f << std::endl;
+  }
+  return res;
+}
+
+int WriteResult(const std::string& imageFile, const std::vector<MSTensor> &outputs) {
+  std::string homePath = "./result_Files";
+  for (size_t i = 0; i < outputs.size(); ++i) {
+    size_t outputSize;
+    std::shared_ptr<const void> netOutput;
+    netOutput = outputs[i].Data();
+    outputSize = outputs[i].DataSize();
+    int pos = imageFile.rfind('/');
+    std::string fileName(imageFile, pos + 1);
+    fileName.replace(fileName.find('.'), fileName.size() - fileName.find('.'), '_' + std::to_string(i) + ".bin");
+    std::string outFileName = homePath + "/" + fileName;
+    FILE * outputFile = fopen(outFileName.c_str(), "wb");
+    fwrite(netOutput.get(), outputSize, sizeof(char), outputFile);
+    fclose(outputFile);
+    outputFile = nullptr;
+  }
+  return 0;
+}
+
+mindspore::MSTensor ReadFileToTensor(const std::string &file) {
+  if (file.empty()) {
+    std::cout << "Pointer file is nullptr" << std::endl;
+    return mindspore::MSTensor();
+  }
+
+  std::ifstream ifs(file);
+  if (!ifs.good()) {
+    std::cout << "File: " << file << " is not exist" << std::endl;
+    return mindspore::MSTensor();
+  }
+
+  if (!ifs.is_open()) {
+    std::cout << "File: " << file << "open failed" << std::endl;
+    return mindspore::MSTensor();
+  }
+
+  ifs.seekg(0, std::ios::end);
+  size_t size = ifs.tellg();
+  mindspore::MSTensor buffer(file, mindspore::DataType::kNumberTypeUInt8, {static_cast<int64_t>(size)}, nullptr, size);
+
+  ifs.seekg(0, std::ios::beg);
+  ifs.read(reinterpret_cast<char *>(buffer.MutableData()), size);
+  ifs.close();
+
+  return buffer;
+}
+
+
+DIR *OpenDir(std::string_view dirName) {
+  if (dirName.empty()) {
+    std::cout << " dirName is null ! " << std::endl;
+    return nullptr;
+  }
+  std::string realPath = RealPath(dirName);
+  struct stat s;
+  lstat(realPath.c_str(), &s);
+  if (!S_ISDIR(s.st_mode)) {
+    std::cout << "dirName is not a valid directory !" << std::endl;
+    return nullptr;
+  }
+  DIR *dir;
+  dir = opendir(realPath.c_str());
+  if (dir == nullptr) {
+    std::cout << "Can not open dir " << dirName << std::endl;
+    return nullptr;
+  }
+  std::cout << "Successfully opened the dir " << dirName << std::endl;
+  return dir;
+}
+
+std::string RealPath(std::string_view path) {
+  char realPathMem[PATH_MAX] = {0};
+  char *realPathRet = nullptr;
+  realPathRet = realpath(path.data(), realPathMem);
+
+  if (realPathRet == nullptr) {
+    std::cout << "File: " << path << " is not exist.";
+    return "";
+  }
+
+  std::string realPath(realPathMem);
+  std::cout << path << " realpath is: " << realPath << std::endl;
+  return realPath;
+}
--- a/model_zoo/official/gnn/bgcf/postprocess.py
+++ b/model_zoo/official/gnn/bgcf/postprocess.py
@ -0,0 +1,75 @@
+# 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.
+# ============================================================================
+"""
+postprocess.
+"""
+import os
+import argparse
+import numpy as np
+
+from src.metrics import BGCFEvaluate
+from src.dataset import load_graph
+
+parser = argparse.ArgumentParser()
+parser.add_argument("--dataset", type=str, default="Beauty", help="choose which dataset")
+parser.add_argument("--datapath", type=str, default="./scripts/data_mr", help="minddata path")
+parser.add_argument('--input_dim', type=int, default=64, choices=[64, 128],
+                    help="user and item embedding dimension")
+parser.add_argument('--Ks', type=list, default=[5, 10, 20, 100], help="top K")
+parser.add_argument('--workers', type=int, default=8, help="number of process to generate data")
+parser.add_argument("--result_path", type=str, default="./result_Files", help="result path")
+args = parser.parse_args()
+
+
+def get_acc():
+    """calculate accuracy"""
+    train_graph, test_graph, _ = load_graph(args.datapath)
+
+    num_user = train_graph.graph_info()["node_num"][0]
+    num_item = train_graph.graph_info()["node_num"][1]
+    input_dim = args.input_dim
+    user_reps = np.zeros([num_user, input_dim * 3])
+    item_reps = np.zeros([num_item, input_dim * 3])
+
+    for i in range(50):
+        sub_folder = os.path.join(args.result_path, 'result_Files_' + str(i))
+        user_rep = np.fromfile(os.path.join(sub_folder, 'amazon-beauty_0.bin'), np.float16)
+        user_rep = user_rep.reshape(num_user, input_dim * 3)
+        item_rep = np.fromfile(os.path.join(sub_folder, 'amazon-beauty_1.bin'), np.float16)
+        item_rep = item_rep.reshape(num_item, input_dim * 3)
+
+        user_reps += user_rep
+        item_reps += item_rep
+    user_reps /= 50
+    item_reps /= 50
+
+    eval_class = BGCFEvaluate(args, train_graph, test_graph, args.Ks)
+
+    test_recall_bgcf, test_ndcg_bgcf, \
+    test_sedp, test_nov = eval_class.eval_with_rep(user_reps, item_reps, args)
+
+    print('recall_@10:%.5f,     recall_@20:%.5f,     ndcg_@10:%.5f,    ndcg_@20:%.5f,   '
+          'sedp_@10:%.5f,     sedp_@20:%.5f,    nov_@10:%.5f,    nov_@20:%.5f\n' % (test_recall_bgcf[1],
+                                                                                    test_recall_bgcf[2],
+                                                                                    test_ndcg_bgcf[1],
+                                                                                    test_ndcg_bgcf[2],
+                                                                                    test_sedp[0],
+                                                                                    test_sedp[1],
+                                                                                    test_nov[1],
+                                                                                    test_nov[2]))
+
+
+if __name__ == "__main__":
+    get_acc()
--- a/model_zoo/official/gnn/bgcf/preprocess.py
+++ b/model_zoo/official/gnn/bgcf/preprocess.py
@ -0,0 +1,90 @@
+# 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.
+# ============================================================================
+"""
+preprocess.
+"""
+import os
+import argparse
+import numpy as np
+
+from mindspore import Tensor
+from mindspore.common import dtype as mstype
+from src.utils import convert_item_id
+from src.dataset import TestGraphDataset, load_graph
+
+parser = argparse.ArgumentParser()
+parser.add_argument("--dataset", type=str, default="Beauty", help="choose which dataset")
+parser.add_argument("--datapath", type=str, default="./scripts/data_mr", help="minddata path")
+parser.add_argument("--num_neg", type=int, default=10, help="negative sampling rate ")
+parser.add_argument("--raw_neighs", type=int, default=40, help="num of sampling neighbors in raw graph")
+parser.add_argument("--gnew_neighs", type=int, default=20, help="num of sampling neighbors in sample graph")
+parser.add_argument("--result_path", type=str, default="./preprocess_Result/", help="result path")
+args = parser.parse_args()
+
+
+
+def get_bin():
+    """generate bin files."""
+    train_graph, _, sampled_graph_list = load_graph(args.datapath)
+    test_graph_dataset = TestGraphDataset(train_graph, sampled_graph_list, num_samples=args.raw_neighs,
+                                          num_bgcn_neigh=args.gnew_neighs,
+                                          num_neg=args.num_neg)
+
+    num_user = train_graph.graph_info()["node_num"][0]
+    num_item = train_graph.graph_info()["node_num"][1]
+
+    for i in range(50):
+        data_path = os.path.join(args.result_path, "data_" + str(i))
+        users_path = os.path.join(data_path, "00_users")
+        os.makedirs(users_path)
+        items_path = os.path.join(data_path, "01_items")
+        os.makedirs(items_path)
+        neg_items_path = os.path.join(data_path, "02_neg_items")
+        os.makedirs(neg_items_path)
+        u_test_neighs_path = os.path.join(data_path, "03_u_test_neighs")
+        os.makedirs(u_test_neighs_path)
+        u_test_gnew_neighs_path = os.path.join(data_path, "04_u_test_gnew_neighs")
+        os.makedirs(u_test_gnew_neighs_path)
+        i_test_neighs_path = os.path.join(data_path, "05_i_test_neighs")
+        os.makedirs(i_test_neighs_path)
+        i_test_gnew_neighs_path = os.path.join(data_path, "06_i_test_gnew_neighs")
+        os.makedirs(i_test_gnew_neighs_path)
+
+        test_graph_dataset.random_select_sampled_graph()
+        u_test_neighs, u_test_gnew_neighs = test_graph_dataset.get_user_sapmled_neighbor()
+        i_test_neighs, i_test_gnew_neighs = test_graph_dataset.get_item_sampled_neighbor()
+
+        u_test_neighs = Tensor(convert_item_id(u_test_neighs, num_user), mstype.int32)
+        u_test_gnew_neighs = Tensor(convert_item_id(u_test_gnew_neighs, num_user), mstype.int32)
+        i_test_neighs = Tensor(i_test_neighs, mstype.int32)
+        i_test_gnew_neighs = Tensor(i_test_gnew_neighs, mstype.int32)
+
+        users = Tensor(np.arange(num_user).reshape(-1,), mstype.int32)
+        items = Tensor(np.arange(num_item).reshape(-1,), mstype.int32)
+        neg_items = Tensor(np.arange(num_item).reshape(-1, 1), mstype.int32)
+
+        file_name = 'amazon-beauty.bin'
+        users.asnumpy().tofile(os.path.join(users_path, file_name))
+        items.asnumpy().tofile(os.path.join(items_path, file_name))
+        neg_items.asnumpy().tofile(os.path.join(neg_items_path, file_name))
+        u_test_neighs.asnumpy().tofile(os.path.join(u_test_neighs_path, file_name))
+        u_test_gnew_neighs.asnumpy().tofile(os.path.join(u_test_gnew_neighs_path, file_name))
+        i_test_neighs.asnumpy().tofile(os.path.join(i_test_neighs_path, file_name))
+        i_test_gnew_neighs.asnumpy().tofile(os.path.join(i_test_gnew_neighs_path, file_name))
+    print("=" * 20, "export bin files finished.", "=" * 20)
+
+
+if __name__ == "__main__":
+    get_bin()
--- a/model_zoo/official/gnn/bgcf/scripts/run_infer_310.sh
+++ b/model_zoo/official/gnn/bgcf/scripts/run_infer_310.sh
@ -0,0 +1,128 @@
+#!/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.
+# ============================================================================
+
+if [[ $# -lt 3 || $# -gt 4 ]]; then
+    echo "Usage: bash run_infer_310.sh [MINDIR_PATH] [DATASET_NAME] [NEED_PREPROCESS] [DEVICE_ID]
+    NEED_PREPROCESS means weather need preprocess or not, it's value is 'y' or 'n'.
+    DEVICE_ID is optional, it can be set by environment variable device_id, otherwise the value is zero"
+exit 1
+fi
+
+get_real_path(){
+    if [ "${1:0:1}" == "/" ]; then
+        echo "$1"
+    else
+        echo "$(realpath -m $PWD/$1)"
+    fi
+}
+model=$(get_real_path $1)
+
+dataset_path=$(get_real_path $2)
+
+if [ "$3" == "y" ] || [ "$3" == "n" ];then
+    need_preprocess=$3
+else
+  echo "weather need preprocess or not, it's value must be in [y, n]"
+  exit 1
+fi
+
+device_id=0
+if [ $# == 4 ]; then
+    device_id=$4
+fi
+
+echo "mindir name: "$model
+echo "dataset path: "$dataset_path
+echo "need preprocess: "$need_preprocess
+echo "device id: "$device_id
+
+export ASCEND_HOME=/usr/local/Ascend/
+if [ -d ${ASCEND_HOME}/ascend-toolkit ]; then
+    export PATH=$ASCEND_HOME/fwkacllib/bin:$ASCEND_HOME/fwkacllib/ccec_compiler/bin:$ASCEND_HOME/ascend-toolkit/latest/fwkacllib/ccec_compiler/bin:$ASCEND_HOME/ascend-toolkit/latest/atc/bin:$PATH
+    export LD_LIBRARY_PATH=$ASCEND_HOME/fwkacllib/lib64:/usr/local/lib:$ASCEND_HOME/ascend-toolkit/latest/atc/lib64:$ASCEND_HOME/ascend-toolkit/latest/fwkacllib/lib64:$ASCEND_HOME/driver/lib64:$ASCEND_HOME/add-ons:$LD_LIBRARY_PATH
+    export TBE_IMPL_PATH=$ASCEND_HOME/ascend-toolkit/latest/opp/op_impl/built-in/ai_core/tbe
+    export PYTHONPATH=$ASCEND_HOME/fwkacllib/python/site-packages:${TBE_IMPL_PATH}:$ASCEND_HOME/ascend-toolkit/latest/fwkacllib/python/site-packages:$PYTHONPATH
+    export ASCEND_OPP_PATH=$ASCEND_HOME/ascend-toolkit/latest/opp
+else
+    export PATH=$ASCEND_HOME/fwkacllib/bin:$ASCEND_HOME/fwkacllib/ccec_compiler/bin:$ASCEND_HOME/atc/ccec_compiler/bin:$ASCEND_HOME/atc/bin:$PATH
+    export LD_LIBRARY_PATH=$ASCEND_HOME/fwkacllib/lib64:/usr/local/lib:$ASCEND_HOME/atc/lib64:$ASCEND_HOME/acllib/lib64:$ASCEND_HOME/driver/lib64:$ASCEND_HOME/add-ons:$LD_LIBRARY_PATH
+    export PYTHONPATH=$ASCEND_HOME/fwkacllib/python/site-packages:$ASCEND_HOME/atc/python/site-packages:$PYTHONPATH
+    export ASCEND_OPP_PATH=$ASCEND_HOME/opp
+fi
+
+function preprocess_data()
+{
+    if [ -d preprocess_Result ]; then
+        rm -rf ./preprocess_Result
+    fi
+    mkdir preprocess_Result
+    python3.7 ../preprocess.py --datapath=$dataset_path --result_path=./preprocess_Result/
+}
+
+function compile_app()
+{
+    cd ../ascend310_infer || exit
+    bash build.sh &> build.log
+}
+
+function infer()
+{
+    cd - || exit
+    if [ -d result_Files ]; then
+        rm -rf ./result_Files
+    fi
+    if [ -d time_Result ]; then
+        rm -rf ./time_Result
+    fi
+    rm -rf ./result_Files_all infer.log
+    mkdir result_Files_all
+    mkdir time_Result
+
+    for i in {0..49}
+    do
+      mkdir result_Files
+      ../ascend310_infer/out/main --mindir_path=$model --input0_path=./preprocess_Result/data_$i/00_users --input1_path=./preprocess_Result/data_$i/01_items --input2_path=./preprocess_Result/data_$i/02_neg_items --input3_path=./preprocess_Result/data_$i/03_u_test_neighs --input4_path=./preprocess_Result/data_$i/04_u_test_gnew_neighs --input5_path=./preprocess_Result/data_$i/05_i_test_neighs --input6_path=./preprocess_Result/data_$i/06_i_test_gnew_neighs --device_id=$device_id &>> infer.log
+      mv result_Files result_Files_all/result_Files_$i
+    done
+}
+
+function cal_acc()
+{
+    python3.7 ../postprocess.py --result_path=./result_Files_all --datapath=$dataset_path &> acc.log
+}
+
+if [ $need_preprocess == "y" ]; then
+    preprocess_data
+    if [ $? -ne 0 ]; then
+        echo "preprocess dataset failed"
+        exit 1
+    fi
+fi
+compile_app
+if [ $? -ne 0 ]; then
+    echo "compile app code failed"
+    exit 1
+fi
+infer
+if [ $? -ne 0 ]; then
+    echo " execute inference failed"
+    exit 1
+fi
+cal_acc
+if [ $? -ne 0 ]; then
+    echo "calculate accuracy failed"
+    exit 1
+fi
--- a/model_zoo/official/nlp/tinybert/README.md
+++ b/model_zoo/official/nlp/tinybert/README.md
@ -26,6 +26,10 @@
            - [evaluation on SST-2 dataset](#evaluation-on-sst-2-dataset)
            - [evaluation on MNLI dataset](#evaluation-on-mnli-dataset)
            - [evaluation on QNLI dataset](#evaluation-on-qnli-dataset)
+    - [Inference Process](#inference-process)
+        - [Export MindIR](#export-mindir)
+        - [Infer on Ascend310](#infer-on-ascend310)
+        - [result](#result)
    - [Model Description](#model-description)
    - [Performance](#performance)
        - [training Performance](#training-performance)
@ -409,6 +413,39 @@ The best acc is 0.891176
 ...
 ```

+## Inference Process
+
+### [Export MindIR](#contents)
+
+```shell
+python export.py --ckpt_file [CKPT_PATH] --file_name [FILE_NAME] --file_format [FILE_FORMAT]
+```
+
+The ckpt_file parameter is required,
+`EXPORT_FORMAT` should be in ["AIR", "MINDIR"]
+
+### Infer on Ascend310
+
+Before performing inference, the mindir file must be exported by `export.py` script. We only provide an example of inference using MINDIR model.
+
+```shell
+# Ascend310 inference
+bash run_infer_310.sh [MINDIR_PATH] [DATASET_PATH] [SCHEMA_DIR] [DATASET_TYPE] [TASK_NAME] [ASSESSMENT_METHOD] [NEED_PREPROCESS] [DEVICE_ID]
+```
+
+- `NEED_PREPROCESS` means weather need preprocess or not, it's value is 'y' or 'n'.
+- `DEVICE_ID` is optional, default value is 0.
+
+### result
+
+Inference result is saved in current path, you can find result like this in acc.log file.
+
+```bash
+=================================================================
+============== acc is 0.8862132352941177
+=================================================================
+```
+
 ## [Model Description](#contents)

 ## [Performance](#contents)
--- a/model_zoo/official/nlp/tinybert/README_CN.md
+++ b/model_zoo/official/nlp/tinybert/README_CN.md
@ -29,6 +29,10 @@
            - [基于SST-2数据集进行评估](#基于sst-2数据集进行评估)
            - [基于MNLI数据集进行评估](#基于mnli数据集进行评估)
            - [基于QNLI数据集进行评估](#基于qnli数据集进行评估)
+    - [推理过程](#推理过程)
+        - [导出MindIR](#导出mindir)
+        - [在Ascend310执行推理](#在ascend310执行推理)
+        - [结果](#结果)
    - [模型描述](#模型描述)
    - [性能](#性能)
        - [评估性能](#评估性能)
@ -410,6 +414,39 @@ The best acc is 0.891176
 ...
 ```

+## 推理过程
+
+### [导出MindIR](#contents)
+
+```shell
+python export.py --ckpt_file [CKPT_PATH] --file_name [FILE_NAME] --file_format [FILE_FORMAT]
+```
+
+参数ckpt_file为必填项，
+`EXPORT_FORMAT` 必须在 ["AIR", "MINDIR"]中选择。
+
+### 在Ascend310执行推理
+
+在执行推理前，mindir文件必须通过`export.py`脚本导出。以下展示了使用minir模型执行推理的示例。
+
+```shell
+# Ascend310 inference
+bash run_infer_310.sh [MINDIR_PATH] [DATASET_PATH] [SCHEMA_DIR] [DATASET_TYPE] [TASK_NAME] [ASSESSMENT_METHOD] [NEED_PREPROCESS] [DEVICE_ID]
+```
+
+- `NEED_PREPROCESS` 表示数据是否需要预处理，取值范围为 'y' 或者 'n'。
+- `DEVICE_ID` 可选，默认值为0。
+
+### 结果
+
+推理结果保存在脚本执行的当前路径，你可以在acc.log中看到以下精度计算结果。
+
+```bash
+=================================================================
+============== acc is 0.8862132352941177
+=================================================================
+```
+
 ## 模型描述

 ## 性能
--- a/model_zoo/official/nlp/tinybert/ascend310_infer/CMakeLists.txt
+++ b/model_zoo/official/nlp/tinybert/ascend310_infer/CMakeLists.txt
@ -0,0 +1,14 @@
+cmake_minimum_required(VERSION 3.14.1)
+project(Ascend310Infer)
+add_compile_definitions(_GLIBCXX_USE_CXX11_ABI=0)
+set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -O0 -g -std=c++17 -Werror -Wall -fPIE -Wl,--allow-shlib-undefined")
+set(PROJECT_SRC_ROOT ${CMAKE_CURRENT_LIST_DIR}/)
+option(MINDSPORE_PATH "mindspore install path" "")
+include_directories(${MINDSPORE_PATH})
+include_directories(${MINDSPORE_PATH}/include)
+include_directories(${PROJECT_SRC_ROOT})
+find_library(MS_LIB libmindspore.so ${MINDSPORE_PATH}/lib)
+file(GLOB_RECURSE MD_LIB ${MINDSPORE_PATH}/_c_dataengine*)
+
+add_executable(main src/main.cc src/utils.cc)
+target_link_libraries(main ${MS_LIB} ${MD_LIB} gflags)
--- a/model_zoo/official/nlp/tinybert/ascend310_infer/build.sh
+++ b/model_zoo/official/nlp/tinybert/ascend310_infer/build.sh
@ -0,0 +1,29 @@
+#!/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.
+# ============================================================================
+if [ -d out ]; then
+    rm -rf out
+fi
+
+mkdir out
+cd out || exit
+
+if [ -f "Makefile" ]; then
+  make clean
+fi
+
+cmake .. \
+    -DMINDSPORE_PATH="`pip3.7 show mindspore-ascend | grep Location | awk '{print $2"/mindspore"}' | xargs realpath`"
+make
--- a/model_zoo/official/nlp/tinybert/ascend310_infer/inc/utils.h
+++ b/model_zoo/official/nlp/tinybert/ascend310_infer/inc/utils.h
@ -0,0 +1,32 @@
+/**
+ * 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.
+ */
+
+#ifndef MINDSPORE_INFERENCE_UTILS_H_
+#define MINDSPORE_INFERENCE_UTILS_H_
+
+#include <sys/stat.h>
+#include <dirent.h>
+#include <vector>
+#include <string>
+#include <memory>
+#include "include/api/types.h"
+
+std::vector<std::string> GetAllFiles(std::string_view dirName);
+DIR *OpenDir(std::string_view dirName);
+std::string RealPath(std::string_view path);
+mindspore::MSTensor ReadFileToTensor(const std::string &file);
+int WriteResult(const std::string& imageFile, const std::vector<mindspore::MSTensor> &outputs);
+#endif
--- a/model_zoo/official/nlp/tinybert/ascend310_infer/src/main.cc
+++ b/model_zoo/official/nlp/tinybert/ascend310_infer/src/main.cc
@ -0,0 +1,142 @@
+/**
+ * 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 <sys/time.h>
+#include <gflags/gflags.h>
+#include <dirent.h>
+#include <iostream>
+#include <string>
+#include <algorithm>
+#include <iosfwd>
+#include <vector>
+#include <fstream>
+#include <sstream>
+
+#include "include/api/model.h"
+#include "include/api/context.h"
+#include "include/api/types.h"
+#include "include/api/serialization.h"
+#include "include/dataset/execute.h"
+#include "include/dataset/vision.h"
+#include "inc/utils.h"
+
+using mindspore::Context;
+using mindspore::Serialization;
+using mindspore::Model;
+using mindspore::Status;
+using mindspore::MSTensor;
+using mindspore::dataset::Execute;
+using mindspore::ModelType;
+using mindspore::GraphCell;
+using mindspore::kSuccess;
+
+DEFINE_string(mindir_path, "", "mindir path");
+DEFINE_string(input0_path, ".", "input0 path");
+DEFINE_string(input1_path, ".", "input1 path");
+DEFINE_string(input2_path, ".", "input2 path");
+DEFINE_int32(device_id, 0, "device id");
+
+int main(int argc, char **argv) {
+  gflags::ParseCommandLineFlags(&argc, &argv, true);
+  if (RealPath(FLAGS_mindir_path).empty()) {
+    std::cout << "Invalid mindir" << std::endl;
+    return 1;
+  }
+
+  auto context = std::make_shared<Context>();
+  auto ascend310 = std::make_shared<mindspore::Ascend310DeviceInfo>();
+  ascend310->SetDeviceID(FLAGS_device_id);
+  ascend310->SetPrecisionMode("allow_fp32_to_fp16");
+  ascend310->SetOpSelectImplMode("high_precision");
+  context->MutableDeviceInfo().push_back(ascend310);
+  mindspore::Graph graph;
+  Serialization::Load(FLAGS_mindir_path, ModelType::kMindIR, &graph);
+
+  Model model;
+  Status ret = model.Build(GraphCell(graph), context);
+  if (ret != kSuccess) {
+    std::cout << "ERROR: Build failed." << std::endl;
+    return 1;
+  }
+
+  std::vector<MSTensor> model_inputs = model.GetInputs();
+  if (model_inputs.empty()) {
+    std::cout << "Invalid model, inputs is empty." << std::endl;
+    return 1;
+  }
+
+  auto input0_files = GetAllFiles(FLAGS_input0_path);
+  auto input1_files = GetAllFiles(FLAGS_input1_path);
+  auto input2_files = GetAllFiles(FLAGS_input2_path);
+
+  if (input0_files.empty() || input1_files.empty() || input2_files.empty()) {
+    std::cout << "ERROR: input data empty." << std::endl;
+    return 1;
+  }
+
+  std::map<double, double> costTime_map;
+  size_t size = input0_files.size();
+
+  for (size_t i = 0; i < size; ++i) {
+    struct timeval start = {0};
+    struct timeval end = {0};
+    double startTimeMs;
+    double endTimeMs;
+    std::vector<MSTensor> inputs;
+    std::vector<MSTensor> outputs;
+    std::cout << "Start predict input files:" << input0_files[i] << std::endl;
+
+    auto input0 = ReadFileToTensor(input0_files[i]);
+    auto input1 = ReadFileToTensor(input1_files[i]);
+    auto input2 = ReadFileToTensor(input2_files[i]);
+    inputs.emplace_back(model_inputs[0].Name(), model_inputs[0].DataType(), model_inputs[0].Shape(),
+                        input0.Data().get(), input0.DataSize());
+    inputs.emplace_back(model_inputs[1].Name(), model_inputs[1].DataType(), model_inputs[1].Shape(),
+                        input1.Data().get(), input1.DataSize());
+    inputs.emplace_back(model_inputs[2].Name(), model_inputs[2].DataType(), model_inputs[2].Shape(),
+                        input2.Data().get(), input2.DataSize());
+
+    gettimeofday(&start, nullptr);
+    ret = model.Predict(inputs, &outputs);
+    gettimeofday(&end, nullptr);
+    if (ret != kSuccess) {
+      std::cout << "Predict " << input0_files[i] << " failed." << std::endl;
+      return 1;
+    }
+    startTimeMs = (1.0 * start.tv_sec * 1000000 + start.tv_usec) / 1000;
+    endTimeMs = (1.0 * end.tv_sec * 1000000 + end.tv_usec) / 1000;
+    costTime_map.insert(std::pair<double, double>(startTimeMs, endTimeMs));
+    WriteResult(input0_files[i], outputs);
+  }
+  double average = 0.0;
+  int inferCount = 0;
+
+  for (auto iter = costTime_map.begin(); iter != costTime_map.end(); iter++) {
+    double diff = 0.0;
+    diff = iter->second - iter->first;
+    average += diff;
+    inferCount++;
+  }
+  average = average / inferCount;
+  std::stringstream timeCost;
+  timeCost << "NN inference cost average time: "<< average << " ms of infer_count " << inferCount << std::endl;
+  std::cout << "NN inference cost average time: "<< average << "ms of infer_count " << inferCount << std::endl;
+  std::string fileName = "./time_Result" + std::string("/test_perform_static.txt");
+  std::ofstream fileStream(fileName.c_str(), std::ios::trunc);
+  fileStream << timeCost.str();
+  fileStream.close();
+  costTime_map.clear();
+  return 0;
+}
--- a/model_zoo/official/nlp/tinybert/ascend310_infer/src/utils.cc
+++ b/model_zoo/official/nlp/tinybert/ascend310_infer/src/utils.cc
@ -0,0 +1,129 @@
+/**
+ * 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 <fstream>
+#include <algorithm>
+#include <iostream>
+#include "inc/utils.h"
+
+using mindspore::MSTensor;
+using mindspore::DataType;
+
+std::vector<std::string> GetAllFiles(std::string_view dirName) {
+  struct dirent *filename;
+  DIR *dir = OpenDir(dirName);
+  if (dir == nullptr) {
+    return {};
+  }
+  std::vector<std::string> res;
+  while ((filename = readdir(dir)) != nullptr) {
+    std::string dName = std::string(filename->d_name);
+    if (dName == "." || dName == ".." || filename->d_type != DT_REG) {
+      continue;
+    }
+    res.emplace_back(std::string(dirName) + "/" + filename->d_name);
+  }
+  std::sort(res.begin(), res.end());
+  for (auto &f : res) {
+    std::cout << "image file: " << f << std::endl;
+  }
+  return res;
+}
+
+int WriteResult(const std::string& imageFile, const std::vector<MSTensor> &outputs) {
+  std::string homePath = "./result_Files";
+  for (size_t i = 0; i < outputs.size(); ++i) {
+    size_t outputSize;
+    std::shared_ptr<const void> netOutput;
+    netOutput = outputs[i].Data();
+    outputSize = outputs[i].DataSize();
+    int pos = imageFile.rfind('/');
+    std::string fileName(imageFile, pos + 1);
+    fileName.replace(fileName.find('.'), fileName.size() - fileName.find('.'), '_' + std::to_string(i) + ".bin");
+    std::string outFileName = homePath + "/" + fileName;
+    FILE * outputFile = fopen(outFileName.c_str(), "wb");
+    fwrite(netOutput.get(), outputSize, sizeof(char), outputFile);
+    fclose(outputFile);
+    outputFile = nullptr;
+  }
+  return 0;
+}
+
+mindspore::MSTensor ReadFileToTensor(const std::string &file) {
+  if (file.empty()) {
+    std::cout << "Pointer file is nullptr" << std::endl;
+    return mindspore::MSTensor();
+  }
+
+  std::ifstream ifs(file);
+  if (!ifs.good()) {
+    std::cout << "File: " << file << " is not exist" << std::endl;
+    return mindspore::MSTensor();
+  }
+
+  if (!ifs.is_open()) {
+    std::cout << "File: " << file << "open failed" << std::endl;
+    return mindspore::MSTensor();
+  }
+
+  ifs.seekg(0, std::ios::end);
+  size_t size = ifs.tellg();
+  mindspore::MSTensor buffer(file, mindspore::DataType::kNumberTypeUInt8, {static_cast<int64_t>(size)}, nullptr, size);
+
+  ifs.seekg(0, std::ios::beg);
+  ifs.read(reinterpret_cast<char *>(buffer.MutableData()), size);
+  ifs.close();
+
+  return buffer;
+}
+
+
+DIR *OpenDir(std::string_view dirName) {
+  if (dirName.empty()) {
+    std::cout << " dirName is null ! " << std::endl;
+    return nullptr;
+  }
+  std::string realPath = RealPath(dirName);
+  struct stat s;
+  lstat(realPath.c_str(), &s);
+  if (!S_ISDIR(s.st_mode)) {
+    std::cout << "dirName is not a valid directory !" << std::endl;
+    return nullptr;
+  }
+  DIR *dir;
+  dir = opendir(realPath.c_str());
+  if (dir == nullptr) {
+    std::cout << "Can not open dir " << dirName << std::endl;
+    return nullptr;
+  }
+  std::cout << "Successfully opened the dir " << dirName << std::endl;
+  return dir;
+}
+
+std::string RealPath(std::string_view path) {
+  char realPathMem[PATH_MAX] = {0};
+  char *realPathRet = nullptr;
+  realPathRet = realpath(path.data(), realPathMem);
+
+  if (realPathRet == nullptr) {
+    std::cout << "File: " << path << " is not exist.";
+    return "";
+  }
+
+  std::string realPath(realPathMem);
+  std::cout << path << " realpath is: " << realPath << std::endl;
+  return realPath;
+}
--- a/model_zoo/official/nlp/tinybert/postprocess.py
+++ b/model_zoo/official/nlp/tinybert/postprocess.py
@ -0,0 +1,103 @@
+# 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.
+# ============================================================================
+
+"""postprocess"""
+
+import os
+import argparse
+import numpy as np
+from mindspore import Tensor
+from src.assessment_method import Accuracy, F1
+from src.td_config import eval_cfg
+
+parser = argparse.ArgumentParser(description='postprocess')
+parser.add_argument("--task_name", type=str, default="", choices=["SST-2", "QNLI", "MNLI", "TNEWS", "CLUENER"],
+                    help="The name of the task to train.")
+parser.add_argument("--assessment_method", type=str, default="accuracy", choices=["accuracy", "bf1", "mf1"],
+                    help="assessment_method include: [accuracy, bf1, mf1], default is accuracy")
+parser.add_argument("--result_path", type=str, default="./result_Files", help="result path")
+parser.add_argument("--label_path", type=str, default="./preprocess_Result/label_ids.npy", help="label path")
+args_opt = parser.parse_args()
+
+
+DEFAULT_NUM_LABELS = 2
+DEFAULT_SEQ_LENGTH = 128
+task_params = {"SST-2": {"num_labels": 2, "seq_length": 64},
+               "QNLI": {"num_labels": 2, "seq_length": 128},
+               "MNLI": {"num_labels": 3, "seq_length": 128},
+               "TNEWS": {"num_labels": 15, "seq_length": 128},
+               "CLUENER": {"num_labels": 43, "seq_length": 128}}
+
+
+class Task:
+    """
+    Encapsulation class of get the task parameter.
+    """
+    def __init__(self, task_name):
+        self.task_name = task_name
+
+    @property
+    def num_labels(self):
+        if self.task_name in task_params and "num_labels" in task_params[self.task_name]:
+            return task_params[self.task_name]["num_labels"]
+        return DEFAULT_NUM_LABELS
+
+    @property
+    def seq_length(self):
+        if self.task_name in task_params and "seq_length" in task_params[self.task_name]:
+            return task_params[self.task_name]["seq_length"]
+        return DEFAULT_SEQ_LENGTH
+task = Task(args_opt.task_name)
+
+def eval_result_print(assessment_method="accuracy", callback=None):
+    """print eval result"""
+    if assessment_method == "accuracy":
+        print("============== acc is {}".format(callback.acc_num / callback.total_num))
+    elif assessment_method == "bf1":
+        print("Precision {:.6f} ".format(callback.TP / (callback.TP + callback.FP)))
+        print("Recall {:.6f} ".format(callback.TP / (callback.TP + callback.FN)))
+        print("F1 {:.6f} ".format(2 * callback.TP / (2 * callback.TP + callback.FP + callback.FN)))
+    elif assessment_method == "mf1":
+        print("F1 {:.6f} ".format(callback.eval()))
+    else:
+        raise ValueError("Assessment method not supported, support: [accuracy, f1]")
+
+def get_acc():
+    """
+    calculate accuracy
+    """
+    if args_opt.assessment_method == "accuracy":
+        callback = Accuracy()
+    elif args_opt.assessment_method == "bf1":
+        callback = F1(num_labels=task.num_labels)
+    elif args_opt.assessment_method == "mf1":
+        callback = F1(num_labels=task.num_labels, mode="MultiLabel")
+    else:
+        raise ValueError("Assessment method not supported, support: [accuracy, f1]")
+    labels = np.load(args_opt.label_path)
+    file_num = len(os.listdir(args_opt.result_path))
+    for i in range(file_num):
+        f_name = "tinybert_bs" + str(eval_cfg.batch_size) + "_" + str(i) + "_0.bin"
+        logits = np.fromfile(os.path.join(args_opt.result_path, f_name), np.float32)
+        logits = logits.reshape(eval_cfg.batch_size, task.num_labels)
+        label_ids = labels[i]
+        callback.update(Tensor(logits), Tensor(label_ids))
+    print("==============================================================")
+    eval_result_print(args_opt.assessment_method, callback)
+    print("==============================================================")
+
+
+if __name__ == '__main__':
+    get_acc()
--- a/model_zoo/official/nlp/tinybert/preprocess.py
+++ b/model_zoo/official/nlp/tinybert/preprocess.py
@ -0,0 +1,75 @@
+# 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.
+# ============================================================================
+
+"""preprocess"""
+
+import os
+import argparse
+import numpy as np
+from src.td_config import eval_cfg
+from src.dataset import create_tinybert_dataset, DataType
+
+
+parser = argparse.ArgumentParser(description='preprocess')
+parser.add_argument("--eval_data_dir", type=str, default="", help="Data path, it is better to use absolute path")
+parser.add_argument("--schema_dir", type=str, default="", help="Schema path, it is better to use absolute path")
+parser.add_argument("--dataset_type", type=str, default="tfrecord",
+                    help="dataset type tfrecord/mindrecord, default is tfrecord")
+parser.add_argument("--result_path", type=str, default="./preprocess_Result/", help="result path")
+args_opt = parser.parse_args()
+
+if args_opt.dataset_type == "tfrecord":
+    dataset_type = DataType.TFRECORD
+elif args_opt.dataset_type == "mindrecord":
+    dataset_type = DataType.MINDRECORD
+else:
+    raise Exception("dataset format is not supported yet")
+
+def get_bin():
+    """
+    generate bin files.
+    """
+    input_ids_path = os.path.join(args_opt.result_path, "00_input_ids")
+    token_type_id_path = os.path.join(args_opt.result_path, "01_token_type_id")
+    input_mask_path = os.path.join(args_opt.result_path, "02_input_mask")
+    label_ids_path = os.path.join(args_opt.result_path, "label_ids.npy")
+
+    os.makedirs(input_ids_path)
+    os.makedirs(token_type_id_path)
+    os.makedirs(input_mask_path)
+
+    eval_dataset = create_tinybert_dataset('td', batch_size=eval_cfg.batch_size,
+                                           device_num=1, rank=0, do_shuffle="false",
+                                           data_dir=args_opt.eval_data_dir,
+                                           schema_dir=args_opt.schema_dir,
+                                           data_type=dataset_type)
+    columns_list = ["input_ids", "input_mask", "segment_ids", "label_ids"]
+    label_list = []
+    for j, data in enumerate(eval_dataset.create_dict_iterator(output_numpy=True, num_epochs=1)):
+        file_name = "tinybert_bs" + str(eval_cfg.batch_size) + "_" + str(j) + ".bin"
+        input_data = []
+        for i in columns_list:
+            input_data.append(data[i])
+        input_ids, input_mask, token_type_id, label_ids = input_data
+        input_ids.tofile(os.path.join(input_ids_path, file_name))
+        input_mask.tofile(os.path.join(input_mask_path, file_name))
+        token_type_id.tofile(os.path.join(token_type_id_path, file_name))
+        label_list.append(label_ids)
+    np.save(label_ids_path, label_list)
+    print("=" * 20, 'export files finished', "=" * 20)
+
+
+if __name__ == '__main__':
+    get_bin()
--- a/model_zoo/official/nlp/tinybert/scripts/run_infer_310.sh
+++ b/model_zoo/official/nlp/tinybert/scripts/run_infer_310.sh
@ -0,0 +1,131 @@
+#!/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.
+# ============================================================================
+
+if [[ $# -lt 7 || $# -gt 8 ]]; then
+    echo "Usage: bash run_infer_310.sh [MINDIR_PATH] [DATASET_PATH] [SCHEMA_DIR] [DATASET_TYPE] [TASK_NAME] [ASSESSMENT_METHOD] [NEED_PREPROCESS] [DEVICE_ID]
+    NEED_PREPROCESS means weather need preprocess or not, it's value is 'y' or 'n'.
+    DEVICE_ID is optional, it can be set by environment variable device_id, otherwise the value is zero"
+exit 1
+fi
+
+get_real_path(){
+    if [ "${1:0:1}" == "/" ]; then
+        echo "$1"
+    else
+        echo "$(realpath -m $PWD/$1)"
+    fi
+}
+model=$(get_real_path $1)
+dataset_path=$(get_real_path $2)
+schema_dir=$(get_real_path $3)
+dataset_type=$4
+task_name=$5
+assessment_method=$6
+
+
+if [ "$7" == "y" ] || [ "$7" == "n" ];then
+    need_preprocess=$7
+else
+  echo "weather need preprocess or not, it's value must be in [y, n]"
+  exit 1
+fi
+
+device_id=0
+if [ $# == 8 ]; then
+    device_id=$8
+fi
+
+echo "mindir name: "$model
+echo "dataset path: "$dataset_path
+echo "schema dir: "$schema_dir
+echo "dataset_type: "$dataset_type
+echo "task_name: "$task_name
+echo "assessment_method: "$assessment_method
+echo "need preprocess: "$need_preprocess
+echo "device id: "$device_id
+
+export ASCEND_HOME=/usr/local/Ascend/
+if [ -d ${ASCEND_HOME}/ascend-toolkit ]; then
+    export PATH=$ASCEND_HOME/fwkacllib/bin:$ASCEND_HOME/fwkacllib/ccec_compiler/bin:$ASCEND_HOME/ascend-toolkit/latest/fwkacllib/ccec_compiler/bin:$ASCEND_HOME/ascend-toolkit/latest/atc/bin:$PATH
+    export LD_LIBRARY_PATH=$ASCEND_HOME/fwkacllib/lib64:/usr/local/lib:$ASCEND_HOME/ascend-toolkit/latest/atc/lib64:$ASCEND_HOME/ascend-toolkit/latest/fwkacllib/lib64:$ASCEND_HOME/driver/lib64:$ASCEND_HOME/add-ons:$LD_LIBRARY_PATH
+    export TBE_IMPL_PATH=$ASCEND_HOME/ascend-toolkit/latest/opp/op_impl/built-in/ai_core/tbe
+    export PYTHONPATH=$ASCEND_HOME/fwkacllib/python/site-packages:${TBE_IMPL_PATH}:$ASCEND_HOME/ascend-toolkit/latest/fwkacllib/python/site-packages:$PYTHONPATH
+    export ASCEND_OPP_PATH=$ASCEND_HOME/ascend-toolkit/latest/opp
+else
+    export PATH=$ASCEND_HOME/fwkacllib/bin:$ASCEND_HOME/fwkacllib/ccec_compiler/bin:$ASCEND_HOME/atc/ccec_compiler/bin:$ASCEND_HOME/atc/bin:$PATH
+    export LD_LIBRARY_PATH=$ASCEND_HOME/fwkacllib/lib64:/usr/local/lib:$ASCEND_HOME/atc/lib64:$ASCEND_HOME/acllib/lib64:$ASCEND_HOME/driver/lib64:$ASCEND_HOME/add-ons:$LD_LIBRARY_PATH
+    export PYTHONPATH=$ASCEND_HOME/fwkacllib/python/site-packages:$ASCEND_HOME/atc/python/site-packages:$PYTHONPATH
+    export ASCEND_OPP_PATH=$ASCEND_HOME/opp
+fi
+
+function preprocess_data()
+{
+    if [ -d preprocess_Result ]; then
+        rm -rf ./preprocess_Result
+    fi
+    mkdir preprocess_Result
+    python3.7 ../preprocess.py --eval_data_dir=$dataset_path --schema_dir=$schema_dir --dataset_type=$dataset_type --result_path=./preprocess_Result/
+}
+
+function compile_app()
+{
+    cd ../ascend310_infer || exit
+    bash build.sh &> build.log
+}
+
+function infer()
+{
+    cd - || exit
+    if [ -d result_Files ]; then
+        rm -rf ./result_Files
+    fi
+    if [ -d time_Result ]; then
+        rm -rf ./time_Result
+    fi
+    mkdir result_Files
+    mkdir time_Result
+
+    ../ascend310_infer/out/main --mindir_path=$model --input0_path=./preprocess_Result/00_input_ids --input1_path=./preprocess_Result/01_token_type_id --input2_path=./preprocess_Result/02_input_mask --device_id=$device_id &> infer.log
+
+}
+
+function cal_acc()
+{
+    python3.7 ../postprocess.py --task_name=$task_name --assessment_method=$assessment_method --result_path=./result_Files --label_path=./preprocess_Result/label_ids.npy &> acc.log
+}
+
+if [ $need_preprocess == "y" ]; then
+    preprocess_data
+    if [ $? -ne 0 ]; then
+        echo "preprocess dataset failed"
+        exit 1
+    fi
+fi
+compile_app
+if [ $? -ne 0 ]; then
+    echo "compile app code failed"
+    exit 1
+fi
+infer
+if [ $? -ne 0 ]; then
+    echo " execute inference failed"
+    exit 1
+fi
+cal_acc
+if [ $? -ne 0 ]; then
+    echo "calculate accuracy failed"
+    exit 1
+fi
--- a/model_zoo/official/recommend/ncf/README.md
+++ b/model_zoo/official/recommend/ncf/README.md
@ -15,6 +15,10 @@
        - [Distributed Training](#distributed-training)  
    - [Evaluation Process](#evaluation-process)
        - [Evaluation](#evaluation)
+    - [Inference Process](#inference-process)
+        - [Export MindIR](#export-mindir)
+        - [Infer on Ascend310](#infer-on-ascend310)
+        - [result](#result)
 - [Model Description](#model-description)
    - [Performance](#performance)  
        - [Evaluation Performance](#evaluation-performance)
@ -220,6 +224,37 @@ Parameters for both training and evaluation can be set in config.py.
  HR:0.6846,NDCG:0.410
  ```

+## Inference Process
+
+### [Export MindIR](#contents)
+
+```shell
+python export.py --ckpt_file [CKPT_PATH] --file_name [FILE_NAME] --file_format [FILE_FORMAT]
+```
+
+The ckpt_file parameter is required,
+`EXPORT_FORMAT` should be in ["AIR", "MINDIR"]
+
+### Infer on Ascend310
+
+Before performing inference, the mindir file must be exported by `export.py` script. We only provide an example of inference using MINDIR model.
+
+```shell
+# Ascend310 inference
+bash run_infer_310.sh [MINDIR_PATH] [NEED_PREPROCESS] [DEVICE_ID]
+```
+
+- `NEED_PREPROCESS` means weather need preprocess or not, it's value is 'y' or 'n'.
+- `DEVICE_ID` is optional, default value is 0.
+
+### result
+
+Inference result is saved in current path, you can find result like this in acc.log file.
+
+```grep "accuracy: " acc.log
+  HR:0.6846,NDCG:0.410
+```
+
 # [Model Description](#contents)

 ## [Performance](#contents)
--- a/model_zoo/official/recommend/ncf/ascend310_infer/CMakeLists.txt
+++ b/model_zoo/official/recommend/ncf/ascend310_infer/CMakeLists.txt
@ -0,0 +1,14 @@
+cmake_minimum_required(VERSION 3.14.1)
+project(Ascend310Infer)
+add_compile_definitions(_GLIBCXX_USE_CXX11_ABI=0)
+set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -O2 -g -std=c++17 -Werror -Wall -fPIE -Wl,--allow-shlib-undefined")
+set(PROJECT_SRC_ROOT ${CMAKE_CURRENT_LIST_DIR}/)
+option(MINDSPORE_PATH "mindspore install path" "")
+include_directories(${MINDSPORE_PATH})
+include_directories(${MINDSPORE_PATH}/include)
+include_directories(${PROJECT_SRC_ROOT})
+find_library(MS_LIB libmindspore.so ${MINDSPORE_PATH}/lib)
+file(GLOB_RECURSE MD_LIB ${MINDSPORE_PATH}/_c_dataengine*)
+
+add_executable(main src/main.cc src/utils.cc)
+target_link_libraries(main ${MS_LIB} ${MD_LIB} gflags)
--- a/model_zoo/official/recommend/ncf/ascend310_infer/build.sh
+++ b/model_zoo/official/recommend/ncf/ascend310_infer/build.sh
@ -0,0 +1,29 @@
+#!/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.
+# ============================================================================
+if [ -d out ]; then
+    rm -rf out
+fi
+
+mkdir out
+cd out || exit
+
+if [ -f "Makefile" ]; then
+  make clean
+fi
+
+cmake .. \
+    -DMINDSPORE_PATH="`pip3.7 show mindspore-ascend | grep Location | awk '{print $2"/mindspore"}' | xargs realpath`"
+make
--- a/model_zoo/official/recommend/ncf/ascend310_infer/inc/utils.h
+++ b/model_zoo/official/recommend/ncf/ascend310_infer/inc/utils.h
@ -0,0 +1,32 @@
+/**
+ * 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.
+ */
+
+#ifndef MINDSPORE_INFERENCE_UTILS_H_
+#define MINDSPORE_INFERENCE_UTILS_H_
+
+#include <sys/stat.h>
+#include <dirent.h>
+#include <vector>
+#include <string>
+#include <memory>
+#include "include/api/types.h"
+
+std::vector<std::string> GetAllFiles(std::string_view dirName);
+DIR *OpenDir(std::string_view dirName);
+std::string RealPath(std::string_view path);
+mindspore::MSTensor ReadFileToTensor(const std::string &file);
+int WriteResult(const std::string& imageFile, const std::vector<mindspore::MSTensor> &outputs);
+#endif
--- a/model_zoo/official/recommend/ncf/ascend310_infer/src/main.cc
+++ b/model_zoo/official/recommend/ncf/ascend310_infer/src/main.cc
@ -0,0 +1,140 @@
+/**
+ * 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 <sys/time.h>
+#include <gflags/gflags.h>
+#include <dirent.h>
+#include <iostream>
+#include <string>
+#include <algorithm>
+#include <iosfwd>
+#include <vector>
+#include <fstream>
+#include <sstream>
+
+#include "include/api/model.h"
+#include "include/api/context.h"
+#include "include/api/types.h"
+#include "include/api/serialization.h"
+#include "include/dataset/execute.h"
+#include "include/dataset/vision.h"
+#include "inc/utils.h"
+
+using mindspore::Context;
+using mindspore::Serialization;
+using mindspore::Model;
+using mindspore::Status;
+using mindspore::MSTensor;
+using mindspore::dataset::Execute;
+using mindspore::ModelType;
+using mindspore::GraphCell;
+using mindspore::kSuccess;
+
+DEFINE_string(mindir_path, "", "mindir path");
+DEFINE_string(input0_path, ".", "input0 path");
+DEFINE_string(input1_path, ".", "input1 path");
+DEFINE_string(input2_path, ".", "input2 path");
+DEFINE_int32(device_id, 0, "device id");
+
+int main(int argc, char **argv) {
+  gflags::ParseCommandLineFlags(&argc, &argv, true);
+  if (RealPath(FLAGS_mindir_path).empty()) {
+    std::cout << "Invalid mindir" << std::endl;
+    return 1;
+  }
+
+  auto context = std::make_shared<Context>();
+  auto ascend310 = std::make_shared<mindspore::Ascend310DeviceInfo>();
+  ascend310->SetDeviceID(FLAGS_device_id);
+  context->MutableDeviceInfo().push_back(ascend310);
+  mindspore::Graph graph;
+  Serialization::Load(FLAGS_mindir_path, ModelType::kMindIR, &graph);
+
+  Model model;
+  Status ret = model.Build(GraphCell(graph), context);
+  if (ret != kSuccess) {
+    std::cout << "ERROR: Build failed." << std::endl;
+    return 1;
+  }
+
+  std::vector<MSTensor> model_inputs = model.GetInputs();
+  if (model_inputs.empty()) {
+    std::cout << "Invalid model, inputs is empty." << std::endl;
+    return 1;
+  }
+
+  auto input0_files = GetAllFiles(FLAGS_input0_path);
+  auto input1_files = GetAllFiles(FLAGS_input1_path);
+  auto input2_files = GetAllFiles(FLAGS_input2_path);
+
+  if (input0_files.empty() || input1_files.empty() || input2_files.empty()) {
+    std::cout << "ERROR: input data empty." << std::endl;
+    return 1;
+  }
+
+  std::map<double, double> costTime_map;
+  size_t size = input0_files.size();
+
+  for (size_t i = 0; i < size; ++i) {
+    struct timeval start = {0};
+    struct timeval end = {0};
+    double startTimeMs;
+    double endTimeMs;
+    std::vector<MSTensor> inputs;
+    std::vector<MSTensor> outputs;
+    std::cout << "Start predict input files:" << input0_files[i] << std::endl;
+
+    auto input0 = ReadFileToTensor(input0_files[i]);
+    auto input1 = ReadFileToTensor(input1_files[i]);
+    auto input2 = ReadFileToTensor(input2_files[i]);
+    inputs.emplace_back(model_inputs[0].Name(), model_inputs[0].DataType(), model_inputs[0].Shape(),
+                        input0.Data().get(), input0.DataSize());
+    inputs.emplace_back(model_inputs[1].Name(), model_inputs[1].DataType(), model_inputs[1].Shape(),
+                        input1.Data().get(), input1.DataSize());
+    inputs.emplace_back(model_inputs[2].Name(), model_inputs[2].DataType(), model_inputs[2].Shape(),
+                        input2.Data().get(), input2.DataSize());
+
+    gettimeofday(&start, nullptr);
+    ret = model.Predict(inputs, &outputs);
+    gettimeofday(&end, nullptr);
+    if (ret != kSuccess) {
+      std::cout << "Predict " << input0_files[i] << " failed." << std::endl;
+      return 1;
+    }
+    startTimeMs = (1.0 * start.tv_sec * 1000000 + start.tv_usec) / 1000;
+    endTimeMs = (1.0 * end.tv_sec * 1000000 + end.tv_usec) / 1000;
+    costTime_map.insert(std::pair<double, double>(startTimeMs, endTimeMs));
+    WriteResult(input0_files[i], outputs);
+  }
+  double average = 0.0;
+  int inferCount = 0;
+
+  for (auto iter = costTime_map.begin(); iter != costTime_map.end(); iter++) {
+    double diff = 0.0;
+    diff = iter->second - iter->first;
+    average += diff;
+    inferCount++;
+  }
+  average = average / inferCount;
+  std::stringstream timeCost;
+  timeCost << "NN inference cost average time: "<< average << " ms of infer_count " << inferCount << std::endl;
+  std::cout << "NN inference cost average time: "<< average << "ms of infer_count " << inferCount << std::endl;
+  std::string fileName = "./time_Result" + std::string("/test_perform_static.txt");
+  std::ofstream fileStream(fileName.c_str(), std::ios::trunc);
+  fileStream << timeCost.str();
+  fileStream.close();
+  costTime_map.clear();
+  return 0;
+}
--- a/model_zoo/official/recommend/ncf/ascend310_infer/src/utils.cc
+++ b/model_zoo/official/recommend/ncf/ascend310_infer/src/utils.cc
@ -0,0 +1,129 @@
+/**
+ * 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 <fstream>
+#include <algorithm>
+#include <iostream>
+#include "inc/utils.h"
+
+using mindspore::MSTensor;
+using mindspore::DataType;
+
+std::vector<std::string> GetAllFiles(std::string_view dirName) {
+  struct dirent *filename;
+  DIR *dir = OpenDir(dirName);
+  if (dir == nullptr) {
+    return {};
+  }
+  std::vector<std::string> res;
+  while ((filename = readdir(dir)) != nullptr) {
+    std::string dName = std::string(filename->d_name);
+    if (dName == "." || dName == ".." || filename->d_type != DT_REG) {
+      continue;
+    }
+    res.emplace_back(std::string(dirName) + "/" + filename->d_name);
+  }
+  std::sort(res.begin(), res.end());
+  for (auto &f : res) {
+    std::cout << "image file: " << f << std::endl;
+  }
+  return res;
+}
+
+int WriteResult(const std::string& imageFile, const std::vector<MSTensor> &outputs) {
+  std::string homePath = "./result_Files";
+  for (size_t i = 0; i < outputs.size(); ++i) {
+    size_t outputSize;
+    std::shared_ptr<const void> netOutput;
+    netOutput = outputs[i].Data();
+    outputSize = outputs[i].DataSize();
+    int pos = imageFile.rfind('/');
+    std::string fileName(imageFile, pos + 1);
+    fileName.replace(fileName.find('.'), fileName.size() - fileName.find('.'), '_' + std::to_string(i) + ".bin");
+    std::string outFileName = homePath + "/" + fileName;
+    FILE * outputFile = fopen(outFileName.c_str(), "wb");
+    fwrite(netOutput.get(), outputSize, sizeof(char), outputFile);
+    fclose(outputFile);
+    outputFile = nullptr;
+  }
+  return 0;
+}
+
+mindspore::MSTensor ReadFileToTensor(const std::string &file) {
+  if (file.empty()) {
+    std::cout << "Pointer file is nullptr" << std::endl;
+    return mindspore::MSTensor();
+  }
+
+  std::ifstream ifs(file);
+  if (!ifs.good()) {
+    std::cout << "File: " << file << " is not exist" << std::endl;
+    return mindspore::MSTensor();
+  }
+
+  if (!ifs.is_open()) {
+    std::cout << "File: " << file << "open failed" << std::endl;
+    return mindspore::MSTensor();
+  }
+
+  ifs.seekg(0, std::ios::end);
+  size_t size = ifs.tellg();
+  mindspore::MSTensor buffer(file, mindspore::DataType::kNumberTypeUInt8, {static_cast<int64_t>(size)}, nullptr, size);
+
+  ifs.seekg(0, std::ios::beg);
+  ifs.read(reinterpret_cast<char *>(buffer.MutableData()), size);
+  ifs.close();
+
+  return buffer;
+}
+
+
+DIR *OpenDir(std::string_view dirName) {
+  if (dirName.empty()) {
+    std::cout << " dirName is null ! " << std::endl;
+    return nullptr;
+  }
+  std::string realPath = RealPath(dirName);
+  struct stat s;
+  lstat(realPath.c_str(), &s);
+  if (!S_ISDIR(s.st_mode)) {
+    std::cout << "dirName is not a valid directory !" << std::endl;
+    return nullptr;
+  }
+  DIR *dir;
+  dir = opendir(realPath.c_str());
+  if (dir == nullptr) {
+    std::cout << "Can not open dir " << dirName << std::endl;
+    return nullptr;
+  }
+  std::cout << "Successfully opened the dir " << dirName << std::endl;
+  return dir;
+}
+
+std::string RealPath(std::string_view path) {
+  char realPathMem[PATH_MAX] = {0};
+  char *realPathRet = nullptr;
+  realPathRet = realpath(path.data(), realPathMem);
+
+  if (realPathRet == nullptr) {
+    std::cout << "File: " << path << " is not exist.";
+    return "";
+  }
+
+  std::string realPath(realPathMem);
+  std::cout << path << " realpath is: " << realPath << std::endl;
+  return realPath;
+}
--- a/model_zoo/official/recommend/ncf/default_config.yaml
+++ b/model_zoo/official/recommend/ncf/default_config.yaml
@ -28,9 +28,15 @@ checkpoint_file_path: "./checkpoint/NCF-14_19418.ckpt"

 # Export options
 device_id: 0
-ckpt_file: ""
-file_name: ""
-file_format: ""
+ckpt_file: "./checkpoint/NCF-25_19418.ckpt"
+file_name: "ncf"
+file_format: "MINDIR"
+
+# Preprocess options
+pre_result_path: "./preprocess_Result"
+
+# Postprocess options
+post_result_path: "./result_Files"

 ---

@ -52,3 +58,5 @@ num_factors: "The Embedding size of MF model."
 checkpoint_path: "The location of the checkpoint file."
 eval_file_name: "Eval output file."
 checkpoint_file_path: "The location of the checkpoint file."
+pre_result_path: "saving dataset to numpy format path."
+post_result_path: "inference result path."
--- a/model_zoo/official/recommend/ncf/postprocess.py
+++ b/model_zoo/official/recommend/ncf/postprocess.py
@ -0,0 +1,52 @@
+# 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.
+# ============================================================================
+"""postprocess"""
+import os
+import numpy as np
+
+from mindspore import Tensor
+from src.metrics import NCFMetric
+from model_utils.config import config
+
+def get_acc():
+    """calculate accuracy"""
+    if not os.path.exists(config.output_path):
+        os.makedirs(config.output_path)
+
+    ncf_metric = NCFMetric()
+    rst_path = config.post_result_path
+    file_num = len(os.listdir(rst_path))//3
+    bs = config.eval_batch_size
+    for i in range(file_num):
+        indice_name = os.path.join(rst_path, "ncf_bs" + str(bs) + "_" + str(i) + "_0.bin")
+        item_name = os.path.join(rst_path, "ncf_bs" + str(bs) + "_" + str(i) + "_1.bin")
+        weight_name = os.path.join(rst_path, "ncf_bs" + str(bs) + "_" + str(i) + "_2.bin")
+
+        batch_indices = np.fromfile(indice_name, np.int32).reshape(1600, 10)
+        batch_items = np.fromfile(item_name, np.int32).reshape(1600, 100)
+        metric_weights = np.fromfile(weight_name, bool)
+        ncf_metric.update(Tensor(batch_indices), Tensor(batch_items), Tensor(metric_weights))
+
+    out = ncf_metric.eval()
+
+    eval_file_path = os.path.join(config.output_path, config.eval_file_name)
+    eval_file = open(eval_file_path, "a+")
+    eval_file.write("EvalCallBack: HR = {}, NDCG = {}\n".format(out[0], out[1]))
+    eval_file.close()
+    print("EvalCallBack: HR = {}, NDCG = {}".format(out[0], out[1]))
+    print("=" * 100 + "Eval Finish!" + "=" * 100)
+
+if __name__ == '__main__':
+    get_acc()
--- a/model_zoo/official/recommend/ncf/preprocess.py
+++ b/model_zoo/official/recommend/ncf/preprocess.py
@ -0,0 +1,46 @@
+# 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.
+# ============================================================================
+"""preprocess"""
+import os
+
+from src.dataset import create_dataset
+from model_utils.config import config
+
+def get_bin():
+    """generate bin files"""
+    ds_eval, _, _ = create_dataset(test_train=False, data_dir=config.data_path,
+                                   dataset=config.dataset, train_epochs=0,
+                                   eval_batch_size=config.eval_batch_size)
+    bs = config.eval_batch_size
+    user_folder = os.path.join(config.pre_result_path, "00_user")
+    os.makedirs(user_folder)
+    item_folder = os.path.join(config.pre_result_path, "01_item")
+    os.makedirs(item_folder)
+    mask_folder = os.path.join(config.pre_result_path, "02_mask")
+    os.makedirs(mask_folder)
+
+    for i, dataset in enumerate(ds_eval.create_tuple_iterator(output_numpy=True)):
+        users, items, masks = dataset
+        file_name = "ncf_bs" + str(bs) + "_" + str(i) + ".bin"
+        users_path = os.path.join(user_folder, file_name)
+        users.tofile(users_path)
+        items_path = os.path.join(item_folder, file_name)
+        items.tofile(items_path)
+        masks_path = os.path.join(mask_folder, file_name)
+        masks.tofile(masks_path)
+    print("=" * 20, "Export bin files success", "=" * 20)
+
+if __name__ == '__main__':
+    get_bin()
--- a/model_zoo/official/recommend/ncf/scripts/run_infer_310.sh
+++ b/model_zoo/official/recommend/ncf/scripts/run_infer_310.sh
@ -0,0 +1,120 @@
+#!/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.
+# ============================================================================
+
+if [[ $# -lt 2 || $# -gt 3 ]]; then
+    echo "Usage: bash run_infer_310.sh [MINDIR_PATH] [NEED_PREPROCESS] [DEVICE_ID]
+    NEED_PREPROCESS means weather need preprocess or not, it's value is 'y' or 'n'.
+    DEVICE_ID is optional, it can be set by environment variable device_id, otherwise the value is zero"
+exit 1
+fi
+
+get_real_path(){
+    if [ "${1:0:1}" == "/" ]; then
+        echo "$1"
+    else
+        echo "$(realpath -m $PWD/$1)"
+    fi
+}
+model=$(get_real_path $1)
+
+if [ "$2" == "y" ] || [ "$2" == "n" ];then
+    need_preprocess=$2
+else
+  echo "weather need preprocess or not, it's value must be in [y, n]"
+  exit 1
+fi
+
+device_id=0
+if [ $# == 3 ]; then
+    device_id=$3
+fi
+
+echo "mindir name: "$model
+echo "need preprocess: "$need_preprocess
+echo "device id: "$device_id
+
+export ASCEND_HOME=/usr/local/Ascend/
+if [ -d ${ASCEND_HOME}/ascend-toolkit ]; then
+    export PATH=$ASCEND_HOME/ascend-toolkit/latest/fwkacllib/ccec_compiler/bin:$ASCEND_HOME/ascend-toolkit/latest/atc/bin:$PATH
+    export LD_LIBRARY_PATH=/usr/local/lib:$ASCEND_HOME/ascend-toolkit/latest/atc/lib64:$ASCEND_HOME/ascend-toolkit/latest/fwkacllib/lib64:$ASCEND_HOME/driver/lib64:$ASCEND_HOME/add-ons:$LD_LIBRARY_PATH
+    export TBE_IMPL_PATH=$ASCEND_HOME/ascend-toolkit/latest/opp/op_impl/built-in/ai_core/tbe
+    export PYTHONPATH=${TBE_IMPL_PATH}:$ASCEND_HOME/ascend-toolkit/latest/fwkacllib/python/site-packages:$PYTHONPATH
+    export ASCEND_OPP_PATH=$ASCEND_HOME/ascend-toolkit/latest/opp
+else
+    export PATH=$ASCEND_HOME/fwkacllib/bin:$ASCEND_HOME/fwkacllib/ccec_compiler/bin:$ASCEND_HOME/atc/ccec_compiler/bin:$ASCEND_HOME/atc/bin:$PATH
+    export LD_LIBRARY_PATH=$ASCEND_HOME/fwkacllib/lib64:/usr/local/lib:$ASCEND_HOME/atc/lib64:$ASCEND_HOME/acllib/lib64:$ASCEND_HOME/driver/lib64:$ASCEND_HOME/add-ons:$LD_LIBRARY_PATH
+    export PYTHONPATH=$ASCEND_HOME/fwkacllib/python/site-packages:$ASCEND_HOME/atc/python/site-packages:$PYTHONPATH
+    export ASCEND_OPP_PATH=$ASCEND_HOME/opp
+fi
+
+function preprocess_data()
+{
+    if [ -d preprocess_Result ]; then
+        rm -rf ./preprocess_Result
+    fi
+    mkdir preprocess_Result
+    python3.7 ../preprocess.py
+}
+
+function compile_app()
+{
+    cd ../ascend310_infer || exit
+    bash build.sh &> build.log
+}
+
+function infer()
+{
+    cd - || exit
+    if [ -d result_Files ]; then
+        rm -rf ./result_Files
+    fi
+    if [ -d time_Result ]; then
+        rm -rf ./time_Result
+    fi
+    mkdir result_Files
+    mkdir time_Result
+
+    ../ascend310_infer/out/main --mindir_path=$model --input0_path=./preprocess_Result/00_user --input1_path=./preprocess_Result/01_item --input2_path=./preprocess_Result/02_mask --device_id=$device_id &> infer.log
+
+}
+
+function cal_acc()
+{
+    python3.7 ../postprocess.py &> acc.log
+}
+
+if [ $need_preprocess == "y" ]; then
+    preprocess_data
+    if [ $? -ne 0 ]; then
+        echo "preprocess dataset failed"
+        exit 1
+    fi
+fi
+compile_app
+if [ $? -ne 0 ]; then
+    echo "compile app code failed"
+    exit 1
+fi
+infer
+if [ $? -ne 0 ]; then
+    echo " execute inference failed"
+    exit 1
+fi
+cal_acc
+if [ $? -ne 0 ]; then
+    echo "calculate accuracy failed"
+    exit 1
+fi
--- a/model_zoo/research/nlp/textrcnn/ascend310_infer/CMakeLists.txt
+++ b/model_zoo/research/nlp/textrcnn/ascend310_infer/CMakeLists.txt
@ -0,0 +1,14 @@
+cmake_minimum_required(VERSION 3.14.1)
+project(Ascend310Infer)
+add_compile_definitions(_GLIBCXX_USE_CXX11_ABI=0)
+set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -O2 -g -std=c++17 -Werror -Wall -fPIE -Wl,--allow-shlib-undefined")
+set(PROJECT_SRC_ROOT ${CMAKE_CURRENT_LIST_DIR}/)
+option(MINDSPORE_PATH "mindspore install path" "")
+include_directories(${MINDSPORE_PATH})
+include_directories(${MINDSPORE_PATH}/include)
+include_directories(${PROJECT_SRC_ROOT})
+find_library(MS_LIB libmindspore.so ${MINDSPORE_PATH}/lib)
+file(GLOB_RECURSE MD_LIB ${MINDSPORE_PATH}/_c_dataengine*)
+
+add_executable(main src/main.cc src/utils.cc)
+target_link_libraries(main ${MS_LIB} ${MD_LIB} gflags)
--- a/model_zoo/research/nlp/textrcnn/ascend310_infer/build.sh
+++ b/model_zoo/research/nlp/textrcnn/ascend310_infer/build.sh
@ -0,0 +1,29 @@
+#!/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.
+# ============================================================================
+if [ -d out ]; then
+    rm -rf out
+fi
+
+mkdir out
+cd out || exit
+
+if [ -f "Makefile" ]; then
+  make clean
+fi
+
+cmake .. \
+    -DMINDSPORE_PATH="`pip3.7 show mindspore-ascend | grep Location | awk '{print $2"/mindspore"}' | xargs realpath`"
+make
--- a/model_zoo/research/nlp/textrcnn/ascend310_infer/inc/utils.h
+++ b/model_zoo/research/nlp/textrcnn/ascend310_infer/inc/utils.h
@ -0,0 +1,32 @@
+/**
+ * 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.
+ */
+
+#ifndef MINDSPORE_INFERENCE_UTILS_H_
+#define MINDSPORE_INFERENCE_UTILS_H_
+
+#include <sys/stat.h>
+#include <dirent.h>
+#include <vector>
+#include <string>
+#include <memory>
+#include "include/api/types.h"
+
+std::vector<std::string> GetAllFiles(std::string_view dirName);
+DIR *OpenDir(std::string_view dirName);
+std::string RealPath(std::string_view path);
+mindspore::MSTensor ReadFileToTensor(const std::string &file);
+int WriteResult(const std::string& imageFile, const std::vector<mindspore::MSTensor> &outputs);
+#endif
--- a/model_zoo/research/nlp/textrcnn/ascend310_infer/src/main.cc
+++ b/model_zoo/research/nlp/textrcnn/ascend310_infer/src/main.cc
@ -0,0 +1,130 @@
+/**
+ * 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 <sys/time.h>
+#include <gflags/gflags.h>
+#include <dirent.h>
+#include <iostream>
+#include <string>
+#include <algorithm>
+#include <iosfwd>
+#include <vector>
+#include <fstream>
+#include <sstream>
+
+#include "include/api/model.h"
+#include "include/api/context.h"
+#include "include/api/types.h"
+#include "include/api/serialization.h"
+#include "include/dataset/execute.h"
+#include "include/dataset/vision.h"
+#include "inc/utils.h"
+
+using mindspore::Context;
+using mindspore::Serialization;
+using mindspore::Model;
+using mindspore::Status;
+using mindspore::MSTensor;
+using mindspore::dataset::Execute;
+using mindspore::ModelType;
+using mindspore::GraphCell;
+using mindspore::kSuccess;
+
+DEFINE_string(mindir_path, "", "mindir path");
+DEFINE_string(input0_path, ".", "input0 path");
+DEFINE_int32(device_id, 0, "device id");
+
+int main(int argc, char **argv) {
+  gflags::ParseCommandLineFlags(&argc, &argv, true);
+  if (RealPath(FLAGS_mindir_path).empty()) {
+    std::cout << "Invalid mindir" << std::endl;
+    return 1;
+  }
+
+  auto context = std::make_shared<Context>();
+  auto ascend310 = std::make_shared<mindspore::Ascend310DeviceInfo>();
+  ascend310->SetDeviceID(FLAGS_device_id);
+  context->MutableDeviceInfo().push_back(ascend310);
+  mindspore::Graph graph;
+  Serialization::Load(FLAGS_mindir_path, ModelType::kMindIR, &graph);
+
+  Model model;
+  Status ret = model.Build(GraphCell(graph), context);
+  if (ret != kSuccess) {
+    std::cout << "ERROR: Build failed." << std::endl;
+    return 1;
+  }
+
+  std::vector<MSTensor> model_inputs = model.GetInputs();
+  if (model_inputs.empty()) {
+    std::cout << "Invalid model, inputs is empty." << std::endl;
+    return 1;
+  }
+
+  auto input0_files = GetAllFiles(FLAGS_input0_path);
+
+  if (input0_files.empty()) {
+    std::cout << "ERROR: input data empty." << std::endl;
+    return 1;
+  }
+
+  std::map<double, double> costTime_map;
+  size_t size = input0_files.size();
+
+  for (size_t i = 0; i < size; ++i) {
+    struct timeval start = {0};
+    struct timeval end = {0};
+    double startTimeMs;
+    double endTimeMs;
+    std::vector<MSTensor> inputs;
+    std::vector<MSTensor> outputs;
+    std::cout << "Start predict input files:" << input0_files[i] << std::endl;
+
+    auto input0 = ReadFileToTensor(input0_files[i]);
+    inputs.emplace_back(model_inputs[0].Name(), model_inputs[0].DataType(), model_inputs[0].Shape(),
+                        input0.Data().get(), input0.DataSize());
+
+    gettimeofday(&start, nullptr);
+    ret = model.Predict(inputs, &outputs);
+    gettimeofday(&end, nullptr);
+    if (ret != kSuccess) {
+      std::cout << "Predict " << input0_files[i] << " failed." << std::endl;
+      return 1;
+    }
+    startTimeMs = (1.0 * start.tv_sec * 1000000 + start.tv_usec) / 1000;
+    endTimeMs = (1.0 * end.tv_sec * 1000000 + end.tv_usec) / 1000;
+    costTime_map.insert(std::pair<double, double>(startTimeMs, endTimeMs));
+    WriteResult(input0_files[i], outputs);
+  }
+  double average = 0.0;
+  int inferCount = 0;
+
+  for (auto iter = costTime_map.begin(); iter != costTime_map.end(); iter++) {
+    double diff = 0.0;
+    diff = iter->second - iter->first;
+    average += diff;
+    inferCount++;
+  }
+  average = average / inferCount;
+  std::stringstream timeCost;
+  timeCost << "NN inference cost average time: "<< average << " ms of infer_count " << inferCount << std::endl;
+  std::cout << "NN inference cost average time: "<< average << "ms of infer_count " << inferCount << std::endl;
+  std::string fileName = "./time_Result" + std::string("/test_perform_static.txt");
+  std::ofstream fileStream(fileName.c_str(), std::ios::trunc);
+  fileStream << timeCost.str();
+  fileStream.close();
+  costTime_map.clear();
+  return 0;
+}
--- a/model_zoo/research/nlp/textrcnn/ascend310_infer/src/utils.cc
+++ b/model_zoo/research/nlp/textrcnn/ascend310_infer/src/utils.cc
@ -0,0 +1,129 @@
+/**
+ * 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 <fstream>
+#include <algorithm>
+#include <iostream>
+#include "inc/utils.h"
+
+using mindspore::MSTensor;
+using mindspore::DataType;
+
+std::vector<std::string> GetAllFiles(std::string_view dirName) {
+  struct dirent *filename;
+  DIR *dir = OpenDir(dirName);
+  if (dir == nullptr) {
+    return {};
+  }
+  std::vector<std::string> res;
+  while ((filename = readdir(dir)) != nullptr) {
+    std::string dName = std::string(filename->d_name);
+    if (dName == "." || dName == ".." || filename->d_type != DT_REG) {
+      continue;
+    }
+    res.emplace_back(std::string(dirName) + "/" + filename->d_name);
+  }
+  std::sort(res.begin(), res.end());
+  for (auto &f : res) {
+    std::cout << "image file: " << f << std::endl;
+  }
+  return res;
+}
+
+int WriteResult(const std::string& imageFile, const std::vector<MSTensor> &outputs) {
+  std::string homePath = "./result_Files";
+  for (size_t i = 0; i < outputs.size(); ++i) {
+    size_t outputSize;
+    std::shared_ptr<const void> netOutput;
+    netOutput = outputs[i].Data();
+    outputSize = outputs[i].DataSize();
+    int pos = imageFile.rfind('/');
+    std::string fileName(imageFile, pos + 1);
+    fileName.replace(fileName.find('.'), fileName.size() - fileName.find('.'), '_' + std::to_string(i) + ".bin");
+    std::string outFileName = homePath + "/" + fileName;
+    FILE * outputFile = fopen(outFileName.c_str(), "wb");
+    fwrite(netOutput.get(), outputSize, sizeof(char), outputFile);
+    fclose(outputFile);
+    outputFile = nullptr;
+  }
+  return 0;
+}
+
+mindspore::MSTensor ReadFileToTensor(const std::string &file) {
+  if (file.empty()) {
+    std::cout << "Pointer file is nullptr" << std::endl;
+    return mindspore::MSTensor();
+  }
+
+  std::ifstream ifs(file);
+  if (!ifs.good()) {
+    std::cout << "File: " << file << " is not exist" << std::endl;
+    return mindspore::MSTensor();
+  }
+
+  if (!ifs.is_open()) {
+    std::cout << "File: " << file << "open failed" << std::endl;
+    return mindspore::MSTensor();
+  }
+
+  ifs.seekg(0, std::ios::end);
+  size_t size = ifs.tellg();
+  mindspore::MSTensor buffer(file, mindspore::DataType::kNumberTypeUInt8, {static_cast<int64_t>(size)}, nullptr, size);
+
+  ifs.seekg(0, std::ios::beg);
+  ifs.read(reinterpret_cast<char *>(buffer.MutableData()), size);
+  ifs.close();
+
+  return buffer;
+}
+
+
+DIR *OpenDir(std::string_view dirName) {
+  if (dirName.empty()) {
+    std::cout << " dirName is null ! " << std::endl;
+    return nullptr;
+  }
+  std::string realPath = RealPath(dirName);
+  struct stat s;
+  lstat(realPath.c_str(), &s);
+  if (!S_ISDIR(s.st_mode)) {
+    std::cout << "dirName is not a valid directory !" << std::endl;
+    return nullptr;
+  }
+  DIR *dir;
+  dir = opendir(realPath.c_str());
+  if (dir == nullptr) {
+    std::cout << "Can not open dir " << dirName << std::endl;
+    return nullptr;
+  }
+  std::cout << "Successfully opened the dir " << dirName << std::endl;
+  return dir;
+}
+
+std::string RealPath(std::string_view path) {
+  char realPathMem[PATH_MAX] = {0};
+  char *realPathRet = nullptr;
+  realPathRet = realpath(path.data(), realPathMem);
+
+  if (realPathRet == nullptr) {
+    std::cout << "File: " << path << " is not exist.";
+    return "";
+  }
+
+  std::string realPath(realPathMem);
+  std::cout << path << " realpath is: " << realPath << std::endl;
+  return realPath;
+}
--- a/model_zoo/research/nlp/textrcnn/postprocess.py
+++ b/model_zoo/research/nlp/textrcnn/postprocess.py
@ -0,0 +1,44 @@
+# 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.
+# ============================================================================
+"""postprocess"""
+import os
+import argparse
+import numpy as np
+
+from mindspore.nn.metrics import Accuracy
+from src.config import textrcnn_cfg as cfg
+
+parser = argparse.ArgumentParser(description='postprocess')
+parser.add_argument('--label_path', type=str, default="./preprocess_Result/label_ids.npy")
+parser.add_argument('--result_path', type=str, default="./result_Files")
+args = parser.parse_args()
+
+def get_acc():
+    '''calculate accuracy'''
+    metric = Accuracy()
+    metric.clear()
+    label_list = np.load(args.label_path, allow_pickle=True)
+    file_num = len(os.listdir(args.result_path))
+
+    for i in range(file_num):
+        f_name = "textcrnn_bs" + str(cfg.batch_size) + "_" + str(i) + "_0.bin"
+        pred = np.fromfile(os.path.join(args.result_path, f_name), np.float16)
+        pred = pred.reshape(cfg.batch_size, int(pred.shape[0]/cfg.batch_size))
+        metric.update(pred, label_list[i])
+    acc = metric.eval()
+    print("============== Accuracy:{} ==============".format(acc))
+
+if __name__ == '__main__':
+    get_acc()
--- a/model_zoo/research/nlp/textrcnn/preprocess.py
+++ b/model_zoo/research/nlp/textrcnn/preprocess.py
@ -0,0 +1,45 @@
+# 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.
+# ============================================================================
+"""preprocess"""
+import os
+import argparse
+import numpy as np
+
+from src.config import textrcnn_cfg as cfg
+from src.dataset import create_dataset
+
+parser = argparse.ArgumentParser(description='preprocess')
+parser.add_argument('--result_path', type=str, default="./preprocess_Result")
+args = parser.parse_args()
+
+def get_bin():
+    '''generate bin files.'''
+    ds_eval = create_dataset(cfg.preprocess_path, cfg.batch_size, False)
+    img_path = os.path.join(args.result_path, "00_feature")
+    os.makedirs(img_path)
+    label_list = []
+
+    for i, data in enumerate(ds_eval.create_dict_iterator(output_numpy=True)):
+        file_name = "textcrnn_bs" + str(cfg.batch_size) + "_" + str(i) + ".bin"
+        file_path = os.path.join(img_path, file_name)
+
+        data["feature"].tofile(file_path)
+        label_list.append(data["label"])
+
+    np.save(os.path.join(args.result_path, "label_ids.npy"), label_list)
+    print("=" * 20, "bin files finished", "=" * 20)
+
+if __name__ == '__main__':
+    get_bin()
--- a/model_zoo/research/nlp/textrcnn/readme.md
+++ b/model_zoo/research/nlp/textrcnn/readme.md
@ -8,6 +8,10 @@
 - [Environment Requirements](#environment-requirements)
 - [Quick Start](#quick-start)
 - [Script Description](#script-description)
+- [Inference Process](#inference-process)
+    - [Export MindIR](#export-mindir)
+    - [Infer on Ascend310](#infer-on-ascend310)
+    - [result](#result)
 - [ModelZoo Homepage](#modelzoo-homepage)

 ## [TextRCNN Description](#contents)
@ -138,6 +142,37 @@ Parameters for both training and evaluation can be set in config.py
 | Accuracy                   | 0.78                      | 0.78 |
 | Speed                      | 35ms/step                  |  77ms/step                         |

+## Inference Process
+
+### [Export MindIR](#contents)
+
+```shell
+python export.py --ckpt_file [CKPT_PATH] --file_name [FILE_NAME] --file_format [FILE_FORMAT]
+```
+
+The ckpt_file parameter is required,
+`EXPORT_FORMAT` should be in ["AIR", "MINDIR"]
+
+### Infer on Ascend310
+
+Before performing inference, the mindir file must be exported by `export.py` script. We only provide an example of inference using MINDIR model.
+
+```shell
+# Ascend310 inference
+bash run_infer_310.sh [MINDIR_PATH] [NEED_PREPROCESS] [DEVICE_ID]
+```
+
+- `NEED_PREPROCESS` means weather need preprocess or not, it's value is 'y' or 'n'.
+- `DEVICE_ID` is optional, default value is 0.
+
+### result
+
+Inference result is saved in current path, you can find result like this in acc.log file.
+
+```bash
+============== Accuracy:{} ============== 0.8008
+```
+
 ## [ModelZoo Homepage](#contents)

 Please check the official [homepage](https://gitee.com/mindspore/mindspore/tree/master/model_zoo).
--- a/model_zoo/research/nlp/textrcnn/scripts/run_infer_310.sh
+++ b/model_zoo/research/nlp/textrcnn/scripts/run_infer_310.sh
@ -0,0 +1,120 @@
+#!/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.
+# ============================================================================
+
+if [[ $# -lt 2 || $# -gt 3 ]]; then
+    echo "Usage: bash run_infer_310.sh [MINDIR_PATH] [NEED_PREPROCESS] [DEVICE_ID]
+    NEED_PREPROCESS means weather need preprocess or not, it's value is 'y' or 'n'.
+    DEVICE_ID is optional, it can be set by environment variable device_id, otherwise the value is zero"
+exit 1
+fi
+
+get_real_path(){
+    if [ "${1:0:1}" == "/" ]; then
+        echo "$1"
+    else
+        echo "$(realpath -m $PWD/$1)"
+    fi
+}
+model=$(get_real_path $1)
+
+if [ "$2" == "y" ] || [ "$2" == "n" ];then
+    need_preprocess=$2
+else
+  echo "weather need preprocess or not, it's value must be in [y, n]"
+  exit 1
+fi
+
+device_id=0
+if [ $# == 3 ]; then
+    device_id=$3
+fi
+
+echo "mindir name: "$model
+echo "need preprocess: "$need_preprocess
+echo "device id: "$device_id
+
+export ASCEND_HOME=/usr/local/Ascend/
+if [ -d ${ASCEND_HOME}/ascend-toolkit ]; then
+    export PATH=$ASCEND_HOME/ascend-toolkit/latest/fwkacllib/ccec_compiler/bin:$ASCEND_HOME/ascend-toolkit/latest/atc/bin:$PATH
+    export LD_LIBRARY_PATH=/usr/local/lib:$ASCEND_HOME/ascend-toolkit/latest/atc/lib64:$ASCEND_HOME/ascend-toolkit/latest/fwkacllib/lib64:$ASCEND_HOME/driver/lib64:$ASCEND_HOME/add-ons:$LD_LIBRARY_PATH
+    export TBE_IMPL_PATH=$ASCEND_HOME/ascend-toolkit/latest/opp/op_impl/built-in/ai_core/tbe
+    export PYTHONPATH=${TBE_IMPL_PATH}:$ASCEND_HOME/ascend-toolkit/latest/fwkacllib/python/site-packages:$PYTHONPATH
+    export ASCEND_OPP_PATH=$ASCEND_HOME/ascend-toolkit/latest/opp
+else
+    export PATH=$ASCEND_HOME/fwkacllib/bin:$ASCEND_HOME/fwkacllib/ccec_compiler/bin:$ASCEND_HOME/atc/ccec_compiler/bin:$ASCEND_HOME/atc/bin:$PATH
+    export LD_LIBRARY_PATH=$ASCEND_HOME/fwkacllib/lib64:/usr/local/lib:$ASCEND_HOME/atc/lib64:$ASCEND_HOME/acllib/lib64:$ASCEND_HOME/driver/lib64:$ASCEND_HOME/add-ons:$LD_LIBRARY_PATH
+    export PYTHONPATH=$ASCEND_HOME/fwkacllib/python/site-packages:$ASCEND_HOME/atc/python/site-packages:$PYTHONPATH
+    export ASCEND_OPP_PATH=$ASCEND_HOME/opp
+fi
+
+function preprocess_data()
+{
+    if [ -d preprocess_Result ]; then
+        rm -rf ./preprocess_Result
+    fi
+    mkdir preprocess_Result
+    python3.7 ../preprocess.py --result_path=./preprocess_Result
+}
+
+function compile_app()
+{
+    cd ../ascend310_infer || exit
+    bash build.sh &> build.log
+}
+
+function infer()
+{
+    cd - || exit
+    if [ -d result_Files ]; then
+        rm -rf ./result_Files
+    fi
+    if [ -d time_Result ]; then
+        rm -rf ./time_Result
+    fi
+    mkdir result_Files
+    mkdir time_Result
+
+    ../ascend310_infer/out/main --mindir_path=$model --input0_path=./preprocess_Result/00_feature --device_id=$device_id &> infer.log
+
+}
+
+function cal_acc()
+{
+    python3.7 ../postprocess.py --label_path=./preprocess_Result/label_ids.npy --result_path=./result_Files &> acc.log
+}
+
+if [ $need_preprocess == "y" ]; then
+    preprocess_data
+    if [ $? -ne 0 ]; then
+        echo "preprocess dataset failed"
+        exit 1
+    fi
+fi
+compile_app
+if [ $? -ne 0 ]; then
+    echo "compile app code failed"
+    exit 1
+fi
+infer
+if [ $? -ne 0 ]; then
+    echo " execute inference failed"
+    exit 1
+fi
+cal_acc
+if [ $? -ne 0 ]; then
+    echo "calculate accuracy failed"
+    exit 1
+fi