diff --git a/python/paddle_serving_server/serve.py b/python/paddle_serving_server/serve.py
index 2fe17970d46a90fb6aa51b69279029cf2e3a4daf..91e4dab3253d35ff14b121e9784ab2fd93044114 100755
--- a/python/paddle_serving_server/serve.py
+++ b/python/paddle_serving_server/serve.py
@@ -99,7 +99,7 @@ def serve_args():
     parser.add_argument(
         "--port", type=int, default=9292, help="Port of the starting gpu")
     parser.add_argument(
-        "--device", type=str, default="gpu", help="Type of device")
+        "--device", type=str, default="cpu", help="Type of device")
     parser.add_argument(
         "--gpu_ids", type=str, default="", nargs="+", help="gpu ids")
     parser.add_argument(
@@ -179,9 +179,9 @@ def serve_args():
 
 def start_gpu_card_model(gpu_mode, port, args):  # pylint: disable=doc-string-missing
 
-    device = "gpu"
-    if gpu_mode == False:
-        device = "cpu"
+    device = "cpu"
+    if gpu_mode == True:
+        device = "gpu"
 
     thread_num = args.thread
     model = args.model
diff --git a/python/paddle_serving_server/web_service.py b/python/paddle_serving_server/web_service.py
index ac161fc1546ccaad5bc7932f99f4577133607643..b2ef5979c2ffd821936655ff2ed5182020b34eb1 100755
--- a/python/paddle_serving_server/web_service.py
+++ b/python/paddle_serving_server/web_service.py
@@ -134,7 +134,7 @@ class WebService(object):
                             op_num=None,
                             op_max_batch=None):
 
-        device = "gpu"
+        device = "cpu"
         server = Server()
         # only when `gpus == None`, which means it`s not set at all
         # we will use the self.gpus.
@@ -149,6 +149,8 @@ class WebService(object):
                 device = "arm"
             else:
                 device = "cpu"
+        else:
+            device = "gpu"
 
         op_maker = OpMaker()
         op_seq_maker = OpSeqMaker()
@@ -227,7 +229,7 @@ class WebService(object):
     def prepare_server(self,
                        workdir,
                        port=9393,
-                       device="gpu",
+                       device="cpu",
                        precision="fp32",
                        use_calib=False,
                        use_lite=False,