diff --git a/research/gnn/dgcn/ascend310_infer/CMakeLists.txt b/research/gnn/dgcn/ascend310_infer/CMakeLists.txt
new file mode 100644
index 0000000000000000000000000000000000000000..435823554c506455be6098283942611ae974f4bf
--- /dev/null
+++ b/research/gnn/dgcn/ascend310_infer/CMakeLists.txt
@@ -0,0 +1,15 @@
+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)
+find_package(gflags REQUIRED)
+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)
diff --git a/research/gnn/dgcn/ascend310_infer/build.sh b/research/gnn/dgcn/ascend310_infer/build.sh
new file mode 100644
index 0000000000000000000000000000000000000000..285514e19f2a1878a7bf8f0eed3c99fbc73868c4
--- /dev/null
+++ b/research/gnn/dgcn/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
diff --git a/research/gnn/dgcn/ascend310_infer/inc/utils.h b/research/gnn/dgcn/ascend310_infer/inc/utils.h
new file mode 100644
index 0000000000000000000000000000000000000000..efebe03a8c1179f5a1f9d5f7ee07e0352a9937c6
--- /dev/null
+++ b/research/gnn/dgcn/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
diff --git a/research/gnn/dgcn/ascend310_infer/src/main.cc b/research/gnn/dgcn/ascend310_infer/src/main.cc
new file mode 100644
index 0000000000000000000000000000000000000000..966bc0d4dd1c826af7017b028c688f449ca64dfa
--- /dev/null
+++ b/research/gnn/dgcn/ascend310_infer/src/main.cc
@@ -0,0 +1,147 @@
+/**
+ * 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;
+  std::cout <<"Start load mindir" << std::endl;
+  Serialization::Load(FLAGS_mindir_path, ModelType::kMindIR, &graph);
+  Model model;
+  std::cout << "Start build graph" << std::endl;
+  Status ret = model.Build(GraphCell(graph), context);
+  if (ret != kSuccess) {
+    std::cout << "ERROR: Build failed." << std::endl;
+    return 1;
+  }
+  std::cout << "Start get inputs" << std::endl;
+  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);
+  std::cout << "size is : " << input0_files.size() << ", " << input1_files.size() << ", "
+            << input2_files.size() << std::endl;
+  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));
+    int ret_ = WriteResult(input0_files[i], outputs);
+    if (ret_ != kSuccess) {
+      std::cout << "write result failed." << std::endl;
+      return 1;
+    }
+  }
+  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;
+}
diff --git a/research/gnn/dgcn/ascend310_infer/src/utils.cc b/research/gnn/dgcn/ascend310_infer/src/utils.cc
new file mode 100644
index 0000000000000000000000000000000000000000..6192ba084f7fe17414c6876f752c80739d98fbef
--- /dev/null
+++ b/research/gnn/dgcn/ascend310_infer/src/utils.cc
@@ -0,0 +1,141 @@
+/**
+ * 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");
+    if (outputFile == nullptr) {
+      std::cout << "open result file" << outFileName << "failed" << std::endl;
+      return -1;
+    }
+    size_t size = fwrite(netOutput.get(), sizeof(char), outputSize, outputFile);
+    if (size != outputSize) {
+      fclose(outputFile);
+      outputFile = nullptr;
+      std::cout << "writer result file" << outFileName << "failed write size[" << size <<
+          "] is smaller than output size[" << outputSize << "], maybe the disk is full" << std::endl;
+      return -1;
+    }
+
+    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;
+}
diff --git a/research/gnn/dgcn/export.py b/research/gnn/dgcn/export.py
index 2991518ef032ae03ec8c3db475ffa40d8ebb7149..8bcc3c8a7842012518462fa300b318a8c345cd8c 100644
--- a/research/gnn/dgcn/export.py
+++ b/research/gnn/dgcn/export.py
@@ -39,21 +39,21 @@ if __name__ == "__main__":
     if args.dataset == "cora":
         input_dim = 1433
         output_dim = 7
-        diffusions = Tensor(np.zeros((2708, 2708), np.float32))
-        ppmi = Tensor(np.zeros((2708, 2708), np.float32))
-        features = Tensor(np.zeros((2708, 1433), np.float32))
+        diffusions = Tensor(np.zeros((2708, 2708), np.float16))
+        ppmi = Tensor(np.zeros((2708, 2708), np.float16))
+        features = Tensor(np.zeros((2708, 1433), np.float16))
     if args.dataset == "citeseer":
         input_dim = 3703
         output_dim = 6
-        diffusions = Tensor(np.zeros((3312, 3312), np.float32))
-        ppmi = Tensor(np.zeros((3312, 3312), np.float32))
-        features = Tensor(np.zeros((3312, 3703), np.float32))
+        diffusions = Tensor(np.zeros((3327, 3327), np.float16))
+        ppmi = Tensor(np.zeros((3327, 3327), np.float16))
+        features = Tensor(np.zeros((3327, 3703), np.float16))
     if args.dataset == "pubmed":
-        input_dim = 3703
+        input_dim = 500
         output_dim = 3
-        diffusions = Tensor(np.zeros((19717, 19717), np.float32))
-        ppmi = Tensor(np.zeros((19717, 19717), np.float32))
-        features = Tensor(np.zeros((19717, 500), np.float32))
+        diffusions = Tensor(np.zeros((19717, 19717), np.float16))
+        ppmi = Tensor(np.zeros((19717, 19717), np.float16))
+        features = Tensor(np.zeros((19717, 500), np.float16))
 
 
     dgcn_net = DGCN(input_dim=input_dim, hidden_dim=config.hidden1, output_dim=output_dim, dropout=config.dropout)
@@ -61,7 +61,8 @@ if __name__ == "__main__":
     dgcn_net.add_flags_recursive(fp16=True)
     param_dict = load_checkpoint(args.ckpt_file)
     load_param_into_net(dgcn_net, param_dict)
-    export(dgcn_net, diffusions, ppmi, features, file_name=args.file_name, file_format=args.file_format)
+    input_data = [diffusions, ppmi, features]
+    export(dgcn_net, *input_data, file_name=args.file_name, file_format=args.file_format)
     print("==========================================")
     print(args.file_name + ".mindir exported successfully!")
     print("==========================================")
diff --git a/research/gnn/dgcn/postprocess.py b/research/gnn/dgcn/postprocess.py
new file mode 100644
index 0000000000000000000000000000000000000000..f60ec1d4af63c5c9167f3dfe1e6b44179893dd48
--- /dev/null
+++ b/research/gnn/dgcn/postprocess.py
@@ -0,0 +1,62 @@
+# 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
+
+def Accuracy(label, mask, preds):
+    """Accuracy with masking."""
+    preds = preds.astype(np.float32)
+    correct_prediction = np.equal(np.argmax(preds, axis=1), np.argmax(label, axis=1))
+    accuracy_all = correct_prediction.astype(np.float32)
+    mask = mask.astype(np.float32)
+    mask_reduce = np.mean(mask)
+    mask = mask / mask_reduce
+    accuracy_all *= mask
+    return np.mean(accuracy_all)
+
+
+def get_acc():
+    """get infer Accuracy."""
+    parser = argparse.ArgumentParser(description='postprocess')
+    parser.add_argument('--dataset_name', type=str, default='cora',
+                        choices=['cora', 'citeseer', 'pubmed'], help='dataset name')
+    parser.add_argument('--result_path', type=str, default='./result_Files', help='result Files')
+    parser.add_argument('--label_path', type=str, default='', help='y_test npy Files')
+    parser.add_argument('--mask_path', type=str, default='', help='test_mask npy Files')
+    args_opt = parser.parse_args()
+
+    label_onehot = np.load(args_opt.label_path)
+    test_mask = np.load(args_opt.mask_path)
+
+    pred = np.fromfile(os.path.join(args_opt.result_path, 'diffusions_0.bin'), np.float16)
+    if args_opt.dataset_name == 'cora':
+        pred = pred.reshape(2708, 7)
+    if args_opt.dataset_name == 'citeseer':
+        pred = pred.reshape(3327, 6)
+    if args_opt.dataset_name == 'pubmed':
+        pred = pred.reshape(19717, 3)
+
+    acc = Accuracy(label_onehot, test_mask, pred)
+    print("Test set results:", "accuracy=", "{:.5f}".format(acc))
+
+if __name__ == '__main__':
+    get_acc()
+    
\ No newline at end of file
diff --git a/research/gnn/dgcn/preprocess.py b/research/gnn/dgcn/preprocess.py
new file mode 100644
index 0000000000000000000000000000000000000000..f1a37c799a144295dc98569aef7d53f9150dc857
--- /dev/null
+++ b/research/gnn/dgcn/preprocess.py
@@ -0,0 +1,62 @@
+# 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.data_process import load_graph_data
+from src.utilities import diffusion_fun_improved_ppmi_dynamic_sparsity, diffusion_fun_sparse
+from src.config import ConfigDGCN
+
+
+def generate_bin():
+    """Generate bin files."""
+    parser = argparse.ArgumentParser(description='preprocess')
+    parser.add_argument('--data_dir', type=str, default='./data/cora/cora_mr', help='Dataset directory')
+    parser.add_argument('--test_nodes_num', type=int, default=1000, help='Nodes numbers for test')
+    parser.add_argument('--result_path', type=str, default='./preprocess_Result/', help='Result path')
+    args_opt = parser.parse_args()
+
+    adj, features, y_train, y_val, y_test, train_mask, val_mask, test_mask, labels = load_graph_data(args_opt.data_dir)
+    print(shape(y_train), shape(y_val), shape(y_test), shape(train_mask), shape(val_mask))
+    adj_path = os.path.join(args_opt.result_path, "00_data")
+    ppmi_path = os.path.join(args_opt.result_path, "01_data")
+    feature_path = os.path.join(args_opt.result_path, "02_data")
+    os.makedirs(adj_path)
+    os.makedirs(feature_path)
+    os.makedirs(ppmi_path)
+    config = ConfigDGCN()
+    diffusions = diffusion_fun_sparse(adj.tocsc())
+    diffusions = diffusions.toarray()
+    ppmi = diffusion_fun_improved_ppmi_dynamic_sparsity(adj, path_len=config.path_len, k=1.0)
+    ppmi = ppmi.toarray()
+    features = features.toarray()
+    diffusions = diffusions.astype(np.float16)
+    ppmi = ppmi.astype(np.float16)
+    features = features.astype(np.float16)
+
+    diffusions.tofile(os.path.join(adj_path, "diffusions.bin"))
+    ppmi.tofile(os.path.join(ppmi_path, "ppmi.bin"))
+    features.tofile(os.path.join(feature_path, "feature.bin"))
+    np.save(os.path.join(args_opt.result_path, 'label_onehot.npy'), labels)
+    np.save(os.path.join(args_opt.result_path, 'test_mask.npy'), test_mask)
+
+if __name__ == '__main__':
+    generate_bin()
+    
\ No newline at end of file
diff --git a/research/gnn/dgcn/readme_CN.md b/research/gnn/dgcn/readme_CN.md
index 27da63d8c0e433120e2d1dade4f350cdc95b2bca..f2f1e6f6824581ce4cf182ea1c70346fb72edf58 100644
--- a/research/gnn/dgcn/readme_CN.md
+++ b/research/gnn/dgcn/readme_CN.md
@@ -14,9 +14,11 @@
             - [鐢ㄦ硶](#鐢ㄦ硶)
             - [鍚姩](#鍚姩)
             - [缁撴灉](#缁撴灉)
-    - [瀵煎嚭MindIR妯″瀷](#瀵煎嚭mindir妯″瀷)
+    - [鎺ㄧ悊杩囩▼](#鎺ㄧ悊杩囩▼)
+       - [瀵煎嚭MindIR](#瀵煎嚭mindir)
+       - [鍦ˋscend310鎵ц鎺ㄧ悊](#鍦ˋscend310鎵ц鎺ㄧ悊)
     - [妯″瀷鎻忚堪](#妯″瀷鎻忚堪)
-        - [鎬ц兘](#鎬ц兘)
+    - [鎬ц兘](#鎬ц兘)
     - [闅忔満鎯呭喌璇存槑](#闅忔満鎯呭喌璇存槑)
 
 <!-- /TOC -->
@@ -235,7 +237,9 @@ Convolution Layers:[(1433, 36), (36, 7)]
 Eval results: loss= 0.52596 accuracy= 0.82800 time= 13.57054
 ```
 
-## [瀵煎嚭MindIR妯″瀷](#contents)
+## 鎺ㄧ悊杩囩▼
+
+### [瀵煎嚭MindIR](#contents)
 
 ```shell
 python export.py --ckpt_file [CKPT_PATH] --file_name [FILE_NAME] --file_format [FILE_FORMAT]
@@ -244,8 +248,29 @@ python export.py --ckpt_file [CKPT_PATH] --file_name [FILE_NAME] --file_format [
 绀轰緥
 
 ```text
-python export.py --ckpt_file ./checkpoint/cora/dgcn.ckpt  --file_format MINDIR
-鍙傛暟ckpt_file涓哄繀濉」锛宍FILE_FORMAT` 蹇呴』鍦� ["AIR", "MINDIR"]涓€夋嫨銆�
+python export.py --ckpt_file ./checkpoint/cora/dgcn.ckpt
+鍙傛暟ckpt_file涓哄繀濉」锛宍EXPORT_FORMAT` 蹇呴』鍦� ["AIR", "MINDIR"]涓€夋嫨銆�
+```
+
+### 鍦ˋscend310鎵ц鎺ㄧ悊
+
+鍦ㄦ墽琛屾帹鐞嗗墠锛宮indir鏂囦欢蹇呴』閫氳繃`export.py`鑴氭湰瀵煎嚭銆備互涓嬪睍绀轰簡浣跨敤minir妯″瀷鎵ц鎺ㄧ悊鐨勭ず渚嬨€�
+
+```shell
+# Ascend310 鎺ㄧ悊
+bash run_infer_310.sh [MINDIR_PATH] [DATASET_NAME] [NEED_PREPROCESS] [DEVICE_ID]
+```
+
+- `DATASET_NAME` 琛ㄧず鏁版嵁闆嗗悕绉帮紝鍙栧€艰寖鍥达細 ['cora', 'citeseer'锛� 'pubmed']銆�
+- `NEED_PREPROCESS` 琛ㄧず鏁版嵁鏄惁闇€瑕侀澶勭悊锛屽彇鍊艰寖鍥达細'y' 鎴栬€� 'n'銆�
+- `DEVICE_ID` 鍙€夛紝榛樿鍊间负0銆�
+
+### result
+
+鎺ㄧ悊缁撴灉淇濆瓨鍦ㄨ剼鏈墽琛岀殑褰撳墠璺緞锛屼綘鍙互鍦╝cc.log涓湅鍒颁互涓嬬簿搴﹁绠楃粨鏋溿€�
+
+```bash
+Test set results: accuracy= 0.82800
 ```
 
 ## 妯″瀷鎻忚堪
diff --git a/research/gnn/dgcn/script/run_eval.sh b/research/gnn/dgcn/script/run_eval.sh
index 9dc7777f2b415714e741de4a4f110556986af65e..4917fa847765ea8dfb2be403b5044290438c09b0 100644
--- a/research/gnn/dgcn/script/run_eval.sh
+++ b/research/gnn/dgcn/script/run_eval.sh
@@ -14,11 +14,14 @@
 # limitations under the License.
 # ============================================================================
 
-if [[ $# -gt 1 ]]; then
-    echo "Usage: bash ./scripts/run_eval.sh [CHECKPOINT]"
+if [[ $# -gt 2 ]]; then
+    echo "Usage: bash run_eval.sh [CHECKPOINT] [DATASET]"
 exit 1
 fi
 
+DATASET_NAME=$2
+CHECKPOINT=$1
+
 if [ ! -d "eval" ]; then
         mkdir eval
 fi
@@ -27,4 +30,4 @@ cp -r ../src ./eval
 cp -r ../data ./eval
 cp -r ../checkpoint ./eval
 cd ./eval || exit
-nohup python -u eval.py --checkpoint=$1 > eval.log 2>&1 &
\ No newline at end of file
+nohup python -u eval.py --checkpoint=$CHECKPOINT --dataset=$DATASET_NAME > eval.log 2>&1 &
\ No newline at end of file
diff --git a/research/gnn/dgcn/script/run_infer_310.sh b/research/gnn/dgcn/script/run_infer_310.sh
new file mode 100644
index 0000000000000000000000000000000000000000..3e40b9bbaefce73e1f62f914cb8c5d3dd756502e
--- /dev/null
+++ b/research/gnn/dgcn/script/run_infer_310.sh
@@ -0,0 +1,129 @@
+#!/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 4 || $# -gt 4 ]]; then
+    echo "Usage: bash run_infer_310.sh [MINDIR_PATH] [DATASET_NAME] [NEED_PREPROCESS] [DEVICE_ID]
+    DATASET_NAME must be in ['cora', 'citeseer', 'pubmed'].
+    NEED_PREPROCESS means weather need preprocess or not, it's value is 'y' or 'n'.
+    DEVICE_ID is optional, it can be set by environment variable device_id, otherwise the value is zero"
+exit 1
+fi
+
+get_real_path(){
+    if [ "${1:0:1}" == "/" ]; then
+        echo "$1"
+    else
+        echo "$(realpath -m $PWD/$1)"
+    fi
+}
+model=$(get_real_path $1)
+if [ "$2" == "cora" ] || [ "$2" == "citeseer" ] || [ "$2" == "pubmed" ];then
+    dataset_name=$2
+else
+  echo "dataset must be in ['cora', 'citeseer', 'pubmed']"
+  exit 1
+fi
+
+
+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 name: "$dataset_name
+echo "need preprocess: "$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=$ASCEND_HOME/opp
+fi
+
+function preprocess_data()
+{
+    if [ -d preprocess_Result ]; then
+        rm -rf ./preprocess_Result
+    fi
+    mkdir preprocess_Result
+    python3.7 ../preprocess.py --data_dir=$dataset_name --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=$model --input0_path=./preprocess_Result/00_data --input1_path=./preprocess_Result/01_data --input2_path=./preprocess_Result/02_data --device_id=$device_id &> infer.log
+
+}
+
+function cal_acc()
+{
+    python3.7 ../postprocess.py --result_path=./result_Files --dataset_name=$dataset_name --label_path=./preprocess_Result/label_onehot.npy --mask_path=./preprocess_Result/test_mask.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
diff --git a/research/gnn/dgcn/script/run_train_8p.sh b/research/gnn/dgcn/script/run_train_8p.sh
index 74ad8123142360aafc4427f716d4d2d2a8b30e7b..872e8ccc94517659ed8abc67c484e649b8fe796f 100644
--- a/research/gnn/dgcn/script/run_train_8p.sh
+++ b/research/gnn/dgcn/script/run_train_8p.sh
@@ -15,7 +15,7 @@
 # ============================================================================
 
 if [[ $# -gt 5 ]]; then
-    echo "Usage: bash ./scripts/run_train_8p.sh [RANK_TABLE] [RANK_SIZE] [DEVICE_START] [DATASET_NAME] [DISTRIBUTED]"
+    echo "Usage: bash run_train_8p.sh [RANK_TABLE] [RANK_SIZE] [DEVICE_START] [DATASET_NAME] [DISTRIBUTED]"
 exit 1
 fi
 
@@ -38,6 +38,7 @@ do
     cp -r ../src ./device$i
     cp -r ../data ./device$i
     cp ../*.py ./device$i
+    cp *.sh ./device$i
     echo "Start training for rank $RANK_ID, device $DEVICE_ID"
     cd ./device$i
     env > env.log
diff --git a/research/gnn/dgcn/train.py b/research/gnn/dgcn/train.py
index 3cc2272585a07716e3af890bb323a65226d3886f..744b26261944feb5a6fd4a5ed5286e616f1b8095 100644
--- a/research/gnn/dgcn/train.py
+++ b/research/gnn/dgcn/train.py
@@ -50,7 +50,7 @@ def run_train(learning_rate=0.01, n_epochs=200, dataset=None, dropout_rate=0.5,
               hidden_size=36):
     """run train."""
     context.set_context(mode=context.GRAPH_MODE, device_target=args.device_target)
-    if args.device_target == "Ascend":
+    if args.device_target == "Ascend" and not args.distributed:
         context.set_context(device_id=args.device_id)
     if args.distributed:
         device_id = int(os.getenv('DEVICE_ID'))