diff --git a/paddle/fluid/inference/io.cc b/paddle/fluid/inference/io.cc
index 65db7c7b5008dcb301e741ec17c3623715e10bb8..98780b6881caa96c0356e07a0a3bd040f28efba0 100644
--- a/paddle/fluid/inference/io.cc
+++ b/paddle/fluid/inference/io.cc
@@ -117,7 +117,7 @@ std::unique_ptr<framework::ProgramDesc> Load(framework::Executor* executor,
   std::string program_desc_str;
   VLOG(3) << "loading model from " << model_filename;
   ReadBinaryFile(model_filename, &program_desc_str);
-
+  // LOG(INFO) << program_desc_str;
   std::unique_ptr<framework::ProgramDesc> main_program(
       new framework::ProgramDesc(program_desc_str));
 
diff --git a/paddle/fluid/inference/tests/book/CMakeLists.txt b/paddle/fluid/inference/tests/book/CMakeLists.txt
index dbb81462b8273bd701e9c9f530eaf69817abd6a1..90357f99d1defb97eaf7137b947d68443ad316a2 100644
--- a/paddle/fluid/inference/tests/book/CMakeLists.txt
+++ b/paddle/fluid/inference/tests/book/CMakeLists.txt
@@ -35,6 +35,7 @@ inference_test(image_classification ARGS vgg resnet)
 inference_test(label_semantic_roles)
 inference_test(recognize_digits ARGS mlp conv)
 inference_test(recommender_system)
+inference_test(nlp)
 #inference_test(rnn_encoder_decoder)
 #inference_test(understand_sentiment ARGS conv)
 inference_test(word2vec)
diff --git a/paddle/fluid/inference/tests/book/test_inference_nlp.cc b/paddle/fluid/inference/tests/book/test_inference_nlp.cc
new file mode 100644
index 0000000000000000000000000000000000000000..0d6d0adfb235d1cf0d39920878f13e3b6c828750
--- /dev/null
+++ b/paddle/fluid/inference/tests/book/test_inference_nlp.cc
@@ -0,0 +1,85 @@
+/* Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
+
+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 "gflags/gflags.h"
+#include "gtest/gtest.h"
+#include "paddle/fluid/inference/tests/test_helper.h"
+
+DEFINE_string(dirname, "", "Directory of the inference model.");
+
+TEST(inference, understand_sentiment) {
+  if (FLAGS_dirname.empty()) {
+    LOG(FATAL) << "Usage: ./example --dirname=path/to/your/model";
+  }
+
+  LOG(INFO) << "FLAGS_dirname: " << FLAGS_dirname << std::endl;
+  std::string dirname = FLAGS_dirname;
+
+  // 0. Call `paddle::framework::InitDevices()` initialize all the devices
+  // In unittests, this is done in paddle/testing/paddle_gtest_main.cc
+  paddle::framework::LoDTensor words;
+
+  paddle::framework::LoD lod{{0, 83}};
+  int64_t word_dict_len = 198392;
+  SetupLoDTensor(&words, lod, static_cast<int64_t>(0),
+                 static_cast<int64_t>(word_dict_len - 1));
+  /*
+    std::vector<int64_t> srcdata{
+        784,    784,   1550,   6463,   56,     75693, 6189,  784,    784,  1550,
+        198391, 6463,  42468,  4376,   10251,  10760, 6189,  297,    396,  6463,
+        6463,   1550,  198391, 6463,   22564,  1612,  291,   68,     164,  784,
+        784,    1550,  198391, 6463,   13659,  3362,  42468, 6189,   2209,
+    198391,
+        6463,   2209,  2209,   198391, 6463,   2209,  1062,  3029,   1831, 3029,
+        1065,   2281,  100,    11216,  1110,   56,    10869, 9811,   100,
+    198391,
+        6463,   100,   9280,   100,    288,    40031, 1680,  1335,   100,  1550,
+        9280,   7265,  244,    1550,   198391, 6463,  1550,  198391, 6463,
+    42468,
+        4376,   10251, 10760};
+    paddle::framework::LoD lod{{0, srcdata.size()}};
+    words.set_lod(lod);
+    int64_t* pdata =
+    words.mutable_data<int64_t>({static_cast<int64_t>(srcdata.size()), 1},
+                                                  paddle::platform::CPUPlace());
+    memcpy(pdata, srcdata.data(), words.numel() * sizeof(int64_t));
+  */
+  LOG(INFO) << "number of input size:" << words.numel();
+  std::vector<paddle::framework::LoDTensor*> cpu_feeds;
+  cpu_feeds.push_back(&words);
+
+  paddle::framework::LoDTensor output1;
+  std::vector<paddle::framework::LoDTensor*> cpu_fetchs1;
+  cpu_fetchs1.push_back(&output1);
+
+  int repeat = 100;
+  // Run inference on CPU
+  TestInference<paddle::platform::CPUPlace, true, true>(dirname, cpu_feeds,
+                                                        cpu_fetchs1, repeat);
+  LOG(INFO) << output1.lod();
+  LOG(INFO) << output1.dims();
+
+#ifdef PADDLE_WITH_CUDA
+  paddle::framework::LoDTensor output2;
+  std::vector<paddle::framework::LoDTensor*> cpu_fetchs2;
+  cpu_fetchs2.push_back(&output2);
+
+  // Run inference on CUDA GPU
+  TestInference<paddle::platform::CUDAPlace>(dirname, cpu_feeds, cpu_fetchs2);
+  LOG(INFO) << output2.lod();
+  LOG(INFO) << output2.dims();
+
+  CheckError<float>(output1, output2);
+#endif
+}
diff --git a/paddle/fluid/inference/tests/test_helper.h b/paddle/fluid/inference/tests/test_helper.h
index 01b8dc0be662da22fe15a79cd9abfe5fa92c9577..1f5551567c6881775c09bd1e52a713261ee87a16 100644
--- a/paddle/fluid/inference/tests/test_helper.h
+++ b/paddle/fluid/inference/tests/test_helper.h
@@ -182,6 +182,9 @@ void TestInference(const std::string& dirname,
         "init_program",
         paddle::platform::DeviceContextPool::Instance().Get(place));
     inference_program = InitProgram(&executor, scope, dirname, is_combined);
+    // std::string binary_str;
+    // inference_program->Proto()->SerializeToString(&binary_str);
+    // LOG(INFO) << binary_str;
     if (use_mkldnn) {
       EnableMKLDNN(inference_program);
     }