diff --git a/paddle/fluid/inference/tests/book/CMakeLists.txt b/paddle/fluid/inference/tests/book/CMakeLists.txt
index 90357f99d1defb97eaf7137b947d68443ad316a2..b33df2942a8262167dee1a022e5dd65eec465523 100644
--- a/paddle/fluid/inference/tests/book/CMakeLists.txt
+++ b/paddle/fluid/inference/tests/book/CMakeLists.txt
@@ -35,7 +35,13 @@ 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)
+
+# This is an unly work around to make this test run
+cc_test(test_inference_nlp
+  SRCS test_inference_nlp.cc
+  DEPS paddle_fluid
+  ARGS
+  --modelpath=${PADDLE_BINARY_DIR}/python/paddle/fluid/tests/book/recognize_digits_mlp.inference.model)
diff --git a/paddle/fluid/inference/tests/book/test_inference_nlp.cc b/paddle/fluid/inference/tests/book/test_inference_nlp.cc
index c4d7b0bbf011f8400ab7e790df1d0217a4a52ab7..5ece6084df272a0db8fc54f6da5206e46e83a821 100644
--- a/paddle/fluid/inference/tests/book/test_inference_nlp.cc
+++ b/paddle/fluid/inference/tests/book/test_inference_nlp.cc
@@ -37,10 +37,22 @@ inline double GetCurrentMs() {
   return 1e+3 * time.tv_sec + 1e-3 * time.tv_usec;
 }
 
+// This function just give dummy data for recognize_digits model.
+size_t DummyData(std::vector<paddle::framework::LoDTensor>* out) {
+  paddle::framework::LoDTensor input;
+  SetupTensor<float>(&input, {1, 1, 28, 28}, -1.f, 1.f);
+  out->emplace_back(input);
+  return 1;
+}
+
 // Load the input word index data from file and save into LodTensor.
 // Return the size of words.
 size_t LoadData(std::vector<paddle::framework::LoDTensor>* out,
                 const std::string& filename) {
+  if (filename.empty()) {
+    return DummyData(out);
+  }
+
   size_t sz = 0;
   std::fstream fin(filename);
   std::string line;
@@ -130,9 +142,12 @@ void ThreadRunInfer(
 }
 
 TEST(inference, nlp) {
-  if (FLAGS_modelpath.empty() || FLAGS_datafile.empty()) {
-    LOG(FATAL) << "Usage: ./example --modelpath=path/to/your/model "
-               << "--datafile=path/to/your/data";
+  if (FLAGS_modelpath.empty()) {
+    LOG(FATAL) << "Usage: ./example --modelpath=path/to/your/model";
+  }
+  if (FLAGS_datafile.empty()) {
+    LOG(WARNING) << " Not data file provided, will use dummy data!"
+                 << "Note: if you use nlp model, please provide data file.";
   }
   LOG(INFO) << "Model Path: " << FLAGS_modelpath;
   LOG(INFO) << "Data File: " << FLAGS_datafile;