diff --git a/python/examples/imdb/clean_data.sh b/python/examples/imdb/clean_data.sh
new file mode 100644
index 0000000000000000000000000000000000000000..6d2c8d7a7ff195f91796483ce8caf9cc5fa0317f
--- /dev/null
+++ b/python/examples/imdb/clean_data.sh
@@ -0,0 +1 @@
+rm -rf imdb.vocab kvdb log *.pyc serving_client_conf serving_server_model test_data text_classification_data.tar.gz train_data work_dir1
diff --git a/python/examples/imdb/local_train.py b/python/examples/imdb/local_train.py
index e2c223afb508f03485cde0c4c206bfb63d6ea601..438fb04f112397ee80feb46540b303535b43054f 100644
--- a/python/examples/imdb/local_train.py
+++ b/python/examples/imdb/local_train.py
@@ -53,7 +53,7 @@ if __name__ == "__main__":
 
     exe = fluid.Executor(fluid.CPUPlace())
     exe.run(fluid.default_startup_program())
-    epochs = 30
+    epochs = 6
     save_dirname = "cnn_model"
 
     import paddle_serving_client.io as serving_io
@@ -62,7 +62,7 @@ if __name__ == "__main__":
         exe.train_from_dataset(program=fluid.default_main_program(),
                                dataset=dataset, debug=False)
         logger.info("TRAIN --> pass: {}".format(i))
-        if i == 20:
+        if i == 5:
             serving_io.save_model("serving_server_model",
                                   "serving_client_conf",
                                   {"words": data, "label": label},
diff --git a/python/examples/imdb/test_client.py b/python/examples/imdb/test_client.py
index 0c9ed0cbd7bca5651358735c5d03893530b4e844..891708db941fce65e2272b60e7520d97723da5ce 100644
--- a/python/examples/imdb/test_client.py
+++ b/python/examples/imdb/test_client.py
@@ -10,8 +10,7 @@ for line in sys.stdin:
     words = [int(x) for x in group[1:int(group[0]) + 1]]
     label = [int(group[-1])]
     feed = {"words": words, "label": label}
-    fetch = ["cost", "acc", "prediction"]
+    fetch = ["acc", "cost", "prediction"]
     fetch_map = client.predict(feed=feed, fetch=fetch)
-    print(fetch_map)
-    #print("{} {}".format(fetch_map["prediction"][1], label[0]))
+    print("{} {}".format(fetch_map["prediction"][1], label[0]))