diff --git a/python/paddle_serving_server/env_check/run.py b/python/paddle_serving_server/env_check/run.py
index 8a304d7889b34b5ed139fde478f186fbbb74a833..6b295554757ed3478bb8a485582a7b43af94b235 100644
--- a/python/paddle_serving_server/env_check/run.py
+++ b/python/paddle_serving_server/env_check/run.py
@@ -1,4 +1,3 @@
-import pytest
 # coding:utf-8
 # Copyright (c) 2020  PaddlePaddle Authors. All Rights Reserved.
 #
@@ -25,7 +24,10 @@ Usage: export PYTHON_EXECUTABLE=/usr/local/bin/python3.6
 
 import sys
 import os
+import pytest
 
+
+inference_test_cases = ["test_fit_a_line.py::TestFitALine::test_inference"]
 cpp_test_cases = ["test_fit_a_line.py::TestFitALine::test_cpu", "test_fit_a_line.py::TestFitALine::test_gpu"]
 pipeline_test_cases = ["test_uci_pipeline.py::TestUCIPipeline::test_cpu", "test_uci_pipeline.py::TestUCIPipeline::test_gpu"]
 
@@ -39,11 +41,16 @@ def run_test_cases(cases_list, case_type):
         args = args_str.split(" ")
         res = pytest.main(args)
         sys.stdout, sys.stderr = old_stdout, old_stderr
+        case_name = case.split('_')[-1]
         if res == 0:
-            print("{} {} environment running success".format(case_type, case[-3:]))
-        else:
-            print("{} {} environment running failure, if you need this environment, please refer to https://github.com/PaddlePaddle/Serving/blob/v0.7.0/doc/Install_CN.md to configure environment".format(case_type, case[-3:]))
-    
+            print("{} {} environment running success".format(case_type, case_name))
+        elif res == 1:
+            if case_name == "inference":
+                print("{} {} environment running failure. Please refer to https://www.paddlepaddle.org.cn/install/quick?docurl=/documentation/docs/zh/install/pip/linux-pip.html to configure environment".format(case_type, case_name))
+                os._exit(0)
+            else:
+                print("{} {} environment running failure, if you need this environment, please refer to https://github.com/PaddlePaddle/Serving/blob/HEAD/doc/Compile_CN.md to configure environment".format(case_type, case_name))
+
 def unset_proxy(key):
     os.unsetenv(key)
 
@@ -51,5 +58,6 @@ def check_env():
     if 'https_proxy' in os.environ or 'http_proxy' in os.environ:
         unset_proxy("https_proxy") 
         unset_proxy("http_proxy")     
+    run_test_cases(inference_test_cases, "PaddlePaddle")
     run_test_cases(cpp_test_cases, "C++")
     run_test_cases(pipeline_test_cases, "Pipeline")
diff --git a/python/paddle_serving_server/env_check/test_fit_a_line.py b/python/paddle_serving_server/env_check/test_fit_a_line.py
index 61064ed45f21c338eb1124d126c0f4aaa64182ff..ad8c21de301424586dbec4366357f910d998ba0e 100644
--- a/python/paddle_serving_server/env_check/test_fit_a_line.py
+++ b/python/paddle_serving_server/env_check/test_fit_a_line.py
@@ -8,10 +8,7 @@ import sys
 
 from paddle_serving_client import Client
 from paddle_serving_client.httpclient import HttpClient
-from paddle_serving_client.io import inference_model_to_serving
-from paddle_serving_app.reader import SegPostprocess
 from paddle_serving_app.reader import *
-import paddle.inference as paddle_infer
 
 from util import *
 
@@ -31,6 +28,11 @@ class TestFitALine(object):
         self.serving_util.release()
 
     def get_truth_val_by_inference(self):
+        try:
+            import paddle.inference as paddle_infer
+        except:
+            # when paddle is not installed, directly return
+            return
         data = np.array(
             [0.0137, -0.1136, 0.2553, -0.0692, 0.0582, -0.0727, -0.1583, -0.0584, 0.6283, 0.4919, 0.1856, 0.0795,
              -0.0332]).astype("float32")[np.newaxis, :]
@@ -55,7 +57,7 @@ class TestFitALine(object):
             output_handle = predictor.get_output_handle(output_data_name)
             output_data = output_handle.copy_to_cpu()
             output_data_dict[output_data_name] = output_data
-        # 对齐Serving output
+        # convert to the same format of Serving output
         print(output_data_dict)
         output_data_dict["price"] = output_data_dict["fc_0.tmp_1"]
         del output_data_dict["fc_0.tmp_1"]
@@ -86,7 +88,12 @@ class TestFitALine(object):
         fetch_map = client.predict(
             feed={"x": data}, fetch=fetch_list, batch=True)
         print(fetch_map)
-        return fetch_map
+        output_dict = self.serving_util.parse_http_result(fetch_map)
+        return output_dict
+
+    def test_inference(self):
+        assert self.truth_val['price'].size != 0
+
 
     def test_cpu(self):
         # 1.start server
@@ -97,9 +104,6 @@ class TestFitALine(object):
 
         # 2.resource check
         assert count_process_num_on_port(9494) == 1
-        # assert check_gpu_memory(0) is False
-
-        # 3.keywords check
 
         # 4.predict by brpc
         # batch_size 1
@@ -120,9 +124,6 @@ class TestFitALine(object):
 
         # 2.resource check
         assert count_process_num_on_port(9494) == 1
-        # assert check_gpu_memory(0) is False
-
-        # 3.keywords check
 
         # 4.predict by brpc 
         # batch_size 1
@@ -134,8 +135,4 @@ class TestFitALine(object):
         # 5.release
         kill_process(9494)
 
-if __name__ == '__main__':
-    sss = TestCPPClient()
-    sss.get_truth_val_by_inference()
-
 
diff --git a/python/paddle_serving_server/env_check/test_uci_pipeline.py b/python/paddle_serving_server/env_check/test_uci_pipeline.py
index d39cf8825fca548dcc114e5b2a5c87fe206df27a..1d32f4e7c7ca6684cc8c73b7bd5d120abfc63c3f 100644
--- a/python/paddle_serving_server/env_check/test_uci_pipeline.py
+++ b/python/paddle_serving_server/env_check/test_uci_pipeline.py
@@ -10,7 +10,6 @@ import sys
 from paddle_serving_server.pipeline import PipelineClient
 from paddle_serving_app.reader import CenterCrop, RGB2BGR, Transpose, Div, Normalize, RCNNPostprocess
 from paddle_serving_app.reader import Sequential, File2Image, Resize, Transpose, BGR2RGB, SegPostprocess
-import paddle.inference as paddle_infer
 
 from util import *
 
@@ -30,6 +29,11 @@ class TestUCIPipeline(object):
         self.serving_util.release()
 
     def get_truth_val_by_inference(self):
+        try:
+            import paddle.inference as paddle_infer
+        except:
+            # when paddle is not installed, directly return
+            return
         data = np.array(
             [0.0137, -0.1136, 0.2553, -0.0692, 0.0582, -0.0727, -0.1583, -0.0584, 0.6283, 0.4919, 0.1856, 0.0795,
              -0.0332]).astype("float32")[np.newaxis, :]
@@ -54,7 +58,7 @@ class TestUCIPipeline(object):
             output_handle = predictor.get_output_handle(output_data_name)
             output_data = output_handle.copy_to_cpu()
             output_data_dict[output_data_name] = output_data
-        # 对齐Serving output
+        # convert to the same format of Serving output
         output_data_dict["prob"] = output_data_dict["fc_0.tmp_1"]
         del output_data_dict["fc_0.tmp_1"]
         self.truth_val = output_data_dict
@@ -63,17 +67,15 @@ class TestUCIPipeline(object):
     def predict_pipeline_rpc(self, batch_size=1):
         # 1.prepare feed_data
         feed_dict = {'x': '0.0137, -0.1136, 0.2553, -0.0692, 0.0582, -0.0727, -0.1583, -0.0584, 0.6283, 0.4919, 0.1856, 0.0795, -0.0332'}
-        # TODO 原示例不支持batch
 
         # 2.init client
-        # fetch = ["label", "prob"]
         client = PipelineClient()
         client.connect(['127.0.0.1:9998'])
 
         # 3.predict for fetch_map
         ret = client.predict(feed_dict=feed_dict)
         print(ret)
-        # 转换为dict
+        # 4.convert dict to numpy
         result = {"prob": np.array(eval(ret.value[0]))}
         print(result)
         return result
@@ -83,7 +85,6 @@ class TestUCIPipeline(object):
         data = '0.0137, -0.1136, 0.2553, -0.0692, 0.0582, -0.0727, -0.1583, -0.0584, 0.6283, 0.4919, 0.1856, 0.0795, ' \
                '-0.0332'
         feed_dict = {"key": [], "value": []}
-        # TODO 原示例不支持batch
         feed_dict["key"].append("x")
         feed_dict["value"].append(data)
 
@@ -91,7 +92,7 @@ class TestUCIPipeline(object):
         url = "http://127.0.0.1:18082/uci/prediction"
         r = requests.post(url=url, data=json.dumps(feed_dict))
         print(r.json())
-        # 转换为dict of numpy array
+        # 3.convert dict to numpy array
         result = {"prob": np.array(eval(r.json()["value"][0]))}
         return result
 
@@ -104,21 +105,19 @@ class TestUCIPipeline(object):
 
         # 2.resource check
         assert count_process_num_on_port(9998) == 1  # gRPC Server
-        assert count_process_num_on_port(18082) == 1  # gRPC gateway 代理、转发
-        #assert check_gpu_memory(0) is False
+        assert count_process_num_on_port(18082) == 1  # gRPC gateway
 
         # 3.keywords check
         check_keywords_in_server_log("MKLDNN is enabled", filename="stderr.log")
 
         # 4.predict by rpc
-        # batch_size=1
         result = self.predict_pipeline_rpc(batch_size=1)
         self.serving_util.check_result(result_data=result, truth_data=self.truth_val, batch_size=1)
-        # # predict by http
-        result = self.predict_pipeline_http(batch_size=1)  # batch_size=1
+        # 5.predict by http
+        result = self.predict_pipeline_http(batch_size=1)  
         self.serving_util.check_result(result_data=result, truth_data=self.truth_val, batch_size=1)
 
-        # 5.release
+        # 6.release
         kill_process(9998)
         kill_process(18082)
 
@@ -132,15 +131,13 @@ class TestUCIPipeline(object):
 
         # 2.resource check
         assert count_process_num_on_port(9998) == 1  # gRPC Server
-        assert count_process_num_on_port(18082) == 1  # gRPC gateway 代理、转发
-        #assert check_gpu_memory(0) is False
+        assert count_process_num_on_port(18082) == 1  # gRPC gateway
 
-        # 4.predict by rpc
-        # batch_size=1
+        # 3.predict by rpc
         result = self.predict_pipeline_rpc(batch_size=1)
         self.serving_util.check_result(result_data=result, truth_data=self.truth_val, batch_size=1)
-        # # predict by http
-        result = self.predict_pipeline_http(batch_size=1)  # batch_size=1
+        # 4.predict by http
+        result = self.predict_pipeline_http(batch_size=1) 
         self.serving_util.check_result(result_data=result, truth_data=self.truth_val, batch_size=1)
 
         # 5.release
diff --git a/python/paddle_serving_server/env_check/util.py b/python/paddle_serving_server/env_check/util.py
index a7a8c1c5a89fc25d36c90db08a42a9f7ba0450e2..ef758c52583753f650e6b18f01c64214fa54aa15 100644
--- a/python/paddle_serving_server/env_check/util.py
+++ b/python/paddle_serving_server/env_check/util.py
@@ -5,7 +5,6 @@ import base64
 import subprocess
 import numpy as np
 
-
 class ServingTest(object):
     def __init__(self, data_path: str, example_path: str, model_dir: str, client_dir: str):
         """
@@ -56,8 +55,6 @@ class ServingTest(object):
             predict_result[key] = value.flatten()
         for key, value in truth_data.items():
             truth_result[key] = np.repeat(value, repeats=batch_size, axis=0).flatten()
-        # print("预测值:", predict_result)
-        # print("真实值:", truth_result)
 
         # compare
         for key in predict_result.keys():
@@ -65,10 +62,6 @@ class ServingTest(object):
             diff_count = np.sum(diff_array > delta)
             assert diff_count == 0, f"total: {np.size(diff_array)} diff count:{diff_count} max:{np.max(diff_array)}"
 
-        # for key in predict_result.keys():
-        #     for i, data in enumerate(predict_result[key]):
-        #         diff = sig_fig_compare(data, truth_result[key][i])
-        #         assert diff < delta, f"data:{data} truth:{truth_result[key][i]} diff is {diff} > {delta}, index:{i}"
 
     @staticmethod
     def parse_http_result(output):
@@ -85,7 +78,6 @@ class ServingTest(object):
 
     @staticmethod
     def release(keywords="web_service.py"):
-        #os.system("kill -9 $(ps -ef | grep serving | awk '{print $2}') > /dev/null 2>&1")
         os.system("kill -9 $(ps -ef | grep " + keywords + " | awk '{print $2}') > /dev/null 2>&1")
 
 
diff --git a/python/paddle_serving_server/serve.py b/python/paddle_serving_server/serve.py
index b2f3dbb3f660f6e433c5db3c329820374840b9fe..6e8cb2832969f05df7e8fbe65c609be15a2d74bd 100755
--- a/python/paddle_serving_server/serve.py
+++ b/python/paddle_serving_server/serve.py
@@ -34,7 +34,7 @@ import socket
 from paddle_serving_server.env import CONF_HOME
 import signal
 from paddle_serving_server.util import *
-from paddle_serving_server.env_check.run import *
+from paddle_serving_server.env_check.run import check_env
 
 
 # web_service.py is still used by Pipeline.
diff --git a/python/requirements.txt b/python/requirements.txt
index c5f3a9248ced0888adfe6b0b5ea82f4a1a07b12f..ddb2e1360527f0b8adb88ff3d347df31ec47af28 100644
--- a/python/requirements.txt
+++ b/python/requirements.txt
@@ -21,4 +21,3 @@ sentencepiece; platform_machine ==  "aarch64"
 opencv-python==4.2.0.32; platform_machine !=  "aarch64"
 opencv-python; platform_machine == "aarch64"
 pytest 
-pynvml
diff --git a/python/setup.py.server.in b/python/setup.py.server.in
index 33d787682ccea77968f2006100b75046e0e92920..4721303e7df86bd293e213ba1fd6da7866f10af0 100644
--- a/python/setup.py.server.in
+++ b/python/setup.py.server.in
@@ -34,7 +34,7 @@ util.gen_pipeline_code("paddle_serving_server")
 REQUIRED_PACKAGES = [
     'six >= 1.10.0', 'protobuf >= 3.11.0', 'grpcio <= 1.33.2', 'grpcio-tools <= 1.33.2',
     'flask >= 1.1.1,<2.0.0', 'click==7.1.2', 'itsdangerous==1.1.0', 'Jinja2==2.11.3',
-    'MarkupSafe==1.1.1', 'Werkzeug==1.0.1', 'func_timeout', 'pyyaml'
+    'MarkupSafe==1.1.1', 'Werkzeug==1.0.1', 'func_timeout', 'pyyaml', 'pytest'
 ]
 
 packages=['paddle_serving_server',