diff --git a/deploy/pdserving/det_local_server.py b/deploy/pdserving/det_local_server.py
index acfccdb6d24a7e1ba490705dd147f21dbf921d31..78970af4d1a8a89f976f48f2c29ec97732afa0ce 100644
--- a/deploy/pdserving/det_local_server.py
+++ b/deploy/pdserving/det_local_server.py
@@ -21,7 +21,10 @@ from paddle_serving_client import Client
 from paddle_serving_app.reader import Sequential, ResizeByFactor
 from paddle_serving_app.reader import Div, Normalize, Transpose
 from paddle_serving_app.reader import DBPostProcess, FilterBoxes
-from paddle_serving_server_gpu.web_service import WebService
+if sys.argv[1] == 'gpu':
+    from paddle_serving_server_gpu.web_service import WebService
+elif sys.argv[1] == 'cpu'
+    from paddle_serving_server.web_service import WebService
 import time
 import re
 import base64
@@ -64,8 +67,11 @@ class OCRService(WebService):
 
 ocr_service = OCRService(name="ocr")
 ocr_service.load_model_config("ocr_det_model")
-ocr_service.set_gpus("0")
-ocr_service.prepare_server(workdir="workdir", port=9292, device="gpu", gpuid=0)
+if sys.argv[1] == 'gpu':
+    ocr_service.set_gpus("0")
+    ocr_service.prepare_server(workdir="workdir", port=9292, device="gpu", gpuid=0)
+elif sys.argv[1] == 'cpu':
+    ocr_service.prepare_server(workdir="workdir", port=9292)
 ocr_service.init_det()
 ocr_service.run_debugger_service()
 ocr_service.run_web_service()
diff --git a/deploy/pdserving/det_web_server.py b/deploy/pdserving/det_web_server.py
index dd69be0c70eb0f4dd627aa47ad33045a204f78c0..14be74130dcb413c31a3e76c150d74f65575f451 100644
--- a/deploy/pdserving/det_web_server.py
+++ b/deploy/pdserving/det_web_server.py
@@ -21,7 +21,10 @@ from paddle_serving_client import Client
 from paddle_serving_app.reader import Sequential, ResizeByFactor
 from paddle_serving_app.reader import Div, Normalize, Transpose
 from paddle_serving_app.reader import DBPostProcess, FilterBoxes
-from paddle_serving_server_gpu.web_service import WebService
+if sys.argv[1] == 'gpu':
+    from paddle_serving_server_gpu.web_service import WebService
+elif sys.argv[1] == 'cpu':
+    from paddle_serving_server.web_service import WebService
 import time
 import re
 import base64
@@ -65,8 +68,11 @@ class OCRService(WebService):
 
 ocr_service = OCRService(name="ocr")
 ocr_service.load_model_config("ocr_det_model")
-ocr_service.set_gpus("0")
-ocr_service.prepare_server(workdir="workdir", port=9292, device="gpu", gpuid=0)
+if sys.argv[1] == 'gpu':
+    ocr_service.set_gpus("0")
+    ocr_service.prepare_server(workdir="workdir", port=9292, device="gpu", gpuid=0)
+elif sys.argv[1] == 'cpu':
+    ocr_service.prepare_server(workdir="workdir", port=9292, device="cpu")
 ocr_service.init_det()
 ocr_service.run_rpc_service()
 ocr_service.run_web_service()
diff --git a/deploy/pdserving/ocr_local_server.py b/deploy/pdserving/ocr_local_server.py
index 93e2d7a3d1dc64451774ecf790c2ebd3b39f1d91..f7458c3036734e4bb6e554097029270e11912a3a 100644
--- a/deploy/pdserving/ocr_local_server.py
+++ b/deploy/pdserving/ocr_local_server.py
@@ -22,7 +22,10 @@ from paddle_serving_client import Client
 from paddle_serving_app.reader import Sequential, URL2Image, ResizeByFactor
 from paddle_serving_app.reader import Div, Normalize, Transpose
 from paddle_serving_app.reader import DBPostProcess, FilterBoxes, GetRotateCropImage, SortedBoxes
-from paddle_serving_server_gpu.web_service import WebService
+if sys.argv[1] == 'gpu':
+    from paddle_serving_server_gpu.web_service import WebService
+elif sys.argv[1] == 'cpu':
+    from paddle_serving_server.web_service import WebService
 from paddle_serving_app.local_predict import Debugger
 import time
 import re
@@ -37,8 +40,12 @@ class OCRService(WebService):
                 (2, 0, 1))
         ])
         self.det_client = Debugger()
-        self.det_client.load_model_config(
-            det_model_config, gpu=True, profile=False)
+        if sys.argv[1] == 'gpu':
+            self.det_client.load_model_config(
+                det_model_config, gpu=True, profile=False)
+        elif sys.argv[1] == 'cpu':
+            self.det_client.load_model_config(
+                det_model_config, gpu=False, profile=False)
         self.ocr_reader = OCRReader()
 
     def preprocess(self, feed=[], fetch=[]):
@@ -99,5 +106,8 @@ ocr_service = OCRService(name="ocr")
 ocr_service.load_model_config("ocr_rec_model")
 ocr_service.prepare_server(workdir="workdir", port=9292)
 ocr_service.init_det_debugger(det_model_config="ocr_det_model")
-ocr_service.run_debugger_service(gpu=True)
+if sys.argv[1] == 'gpu':
+    ocr_service.run_debugger_service(gpu=True)
+elif sys.argv[1] == 'cpu':
+    ocr_service.run_debugger_service()
 ocr_service.run_web_service()
diff --git a/deploy/pdserving/ocr_web_server.py b/deploy/pdserving/ocr_web_server.py
index d017f6b9b560dc82158641b9f3a9f80137b40716..6c0de44661958a6425f57039261969551ff552c5 100644
--- a/deploy/pdserving/ocr_web_server.py
+++ b/deploy/pdserving/ocr_web_server.py
@@ -22,7 +22,10 @@ from paddle_serving_client import Client
 from paddle_serving_app.reader import Sequential, URL2Image, ResizeByFactor
 from paddle_serving_app.reader import Div, Normalize, Transpose
 from paddle_serving_app.reader import DBPostProcess, FilterBoxes, GetRotateCropImage, SortedBoxes
-from paddle_serving_server_gpu.web_service import WebService
+if sys.argv[1] == 'gpu':
+    from paddle_serving_server_gpu.web_service import WebService
+elif sys.argv[1] == 'cpu':
+    from paddle_serving_server.web_service import WebService
 import time
 import re
 import base64
@@ -90,8 +93,11 @@ class OCRService(WebService):
 
 ocr_service = OCRService(name="ocr")
 ocr_service.load_model_config("ocr_rec_model")
-ocr_service.set_gpus("0")
-ocr_service.prepare_server(workdir="workdir", port=9292, device="gpu", gpuid=0)
+if sys.argv[1] == 'gpu':
+    ocr_service.set_gpus("0")
+    ocr_service.prepare_server(workdir="workdir", port=9292, device="gpu", gpuid=0)
+elif sys.argv[1] == 'cpu':
+    ocr_service.prepare_server(workdir="workdir", port=9292)
 ocr_service.init_det_client(
     det_port=9293,
     det_client_config="ocr_det_client/serving_client_conf.prototxt")
diff --git a/deploy/pdserving/readme.md b/deploy/pdserving/readme.md
index d6069cd35d201ee7418324c5d9b0b8b28c6569bd..9472e94cffcd483a85850f6e7ea9c8bc172aaf3b 100644
--- a/deploy/pdserving/readme.md
+++ b/deploy/pdserving/readme.md
@@ -16,20 +16,33 @@
 
 **Python3操作指南：**
 ```
-#以下提供beta版本的paddle serving whl包，欢迎试用，正式版会在7月底正式上线
+#以下提供beta版本的paddle serving whl包，欢迎试用，正式版会在8月中正式上线
+#GPU用户下载server包使用这个链接
 wget --no-check-certificate https://paddle-serving.bj.bcebos.com/others/paddle_serving_server_gpu-0.3.2-py3-none-any.whl
+python -m pip install paddle_serving_server_gpu-0.3.2-py3-none-any.whl
+#CPU版本使用这个链接
+wget --no-check-certificate https://paddle-serving.bj.bcebos.com/others/paddle_serving_server-0.3.2-py3-none-any.whl
+python -m pip install paddle_serving_server-0.3.2-py3-none-any.whl
+#客户端和App包使用以下链接（CPU，GPU通用）
 wget --no-check-certificate https://paddle-serving.bj.bcebos.com/others/paddle_serving_client-0.3.2-cp36-none-any.whl
 wget --no-check-certificate https://paddle-serving.bj.bcebos.com/others/paddle_serving_app-0.1.2-py3-none-any.whl
-python -m pip install paddle_serving_app-0.1.2-py3-none-any.whl paddle_serving_server_gpu-0.3.2-py3-none-any.whl paddle_serving_client-0.3.2-cp36-none-any.whl
+python -m pip install paddle_serving_app-0.1.2-py3-none-any.whl paddle_serving_client-0.3.2-cp36-none-any.whl
 ```
 
 **Python2操作指南：**
 ```
-#以下提供beta版本的paddle serving whl包，欢迎试用，正式版会在7月底正式上线
+#以下提供beta版本的paddle serving whl包，欢迎试用，正式版会在8月中正式上线
+#GPU用户下载server包使用这个链接
 wget --no-check-certificate https://paddle-serving.bj.bcebos.com/others/paddle_serving_server_gpu-0.3.2-py2-none-any.whl
+python -m pip install paddle_serving_server_gpu-0.3.2-py2-none-any.whl 
+#CPU版本使用这个链接
+wget --no-check-certificate https://paddle-serving.bj.bcebos.com/others/paddle_serving_server-0.3.2-py2-none-any.whl
+python -m pip install paddle_serving_server-0.3.2-py2-none-any.whl
+
+#客户端和App包使用以下链接（CPU，GPU通用）
 wget --no-check-certificate https://paddle-serving.bj.bcebos.com/others/paddle_serving_app-0.1.2-py2-none-any.whl
 wget --no-check-certificate https://paddle-serving.bj.bcebos.com/others/paddle_serving_client-0.3.2-cp27-none-any.whl
-python -m pip install paddle_serving_app-0.1.2-py2-none-any.whl paddle_serving_server_gpu-0.3.2-py2-none-any.whl paddle_serving_client-0.3.2-cp27-none-any.whl
+python -m pip install paddle_serving_app-0.1.2-py2-none-any.whl paddle_serving_client-0.3.2-cp27-none-any.whl
 ```
 
 ### 2. 模型转换
@@ -53,14 +66,21 @@ tar -xzvf ocr_det.tar.gz
 #### 方式1. 启动标准版服务
 
 ```
+# cpu，gpu启动二选一，以下是cpu启动
+python -m paddle_serving_server.serve --model ocr_det_model --port 9293 
+python ocr_web_server.py cpu
+# gpu启动
 python -m paddle_serving_server_gpu.serve --model ocr_det_model --port 9293 --gpu_id 0
-python ocr_web_server.py
+python ocr_web_server.py gpu
 ```
 
 #### 方式2. 启动快速版服务
 
 ```
-python ocr_local_server.py
+# cpu，gpu启动二选一，以下是cpu启动
+python ocr_local_server.py cpu
+# gpu启动
+python ocr_local_server.py gpu
 ```
 
 ## 发送预测请求
@@ -85,7 +105,7 @@ python ocr_web_client.py
 
 在`ocr_web_server.py`或是`ocr_local_server.py`当中的`preprocess`函数里面做了检测服务和识别服务的前处理，`postprocess`函数里面做了识别的后处理服务，可以在相应的函数中做修改。调用了`paddle_serving_app`库提供的常见CV模型的前处理/后处理库。
 
-如果想要单独启动Paddle Serving的检测服务和识别服务，参见下列表格, 执行对应的脚本即可。
+如果想要单独启动Paddle Serving的检测服务和识别服务，参见下列表格, 执行对应的脚本即可，并且在命令行参数注明用的CPU或是GPU来提供服务。
 
 | 模型 | 标准版         | 快速版           |
 | ---- | ----------------- | ------------------- |
diff --git a/deploy/pdserving/rec_web_server.py b/deploy/pdserving/rec_web_server.py
index 684c313d4d50cfe00c576c81aad05a810525dcce..0f4e9f6d264ed602f387bfaf0303cd59af7823fa 100644
--- a/deploy/pdserving/rec_web_server.py
+++ b/deploy/pdserving/rec_web_server.py
@@ -22,7 +22,10 @@ from paddle_serving_client import Client
 from paddle_serving_app.reader import Sequential, URL2Image, ResizeByFactor
 from paddle_serving_app.reader import Div, Normalize, Transpose
 from paddle_serving_app.reader import DBPostProcess, FilterBoxes, GetRotateCropImage, SortedBoxes
-from paddle_serving_server_gpu.web_service import WebService
+if sys.argv[1] == 'gpu':
+    from paddle_serving_server_gpu.web_service import WebService
+elif sys.argv[1] == 'cpu':
+    from paddle_serving_server.web_service import WebService
 import time
 import re
 import base64
@@ -64,8 +67,11 @@ class OCRService(WebService):
 
 ocr_service = OCRService(name="ocr")
 ocr_service.load_model_config("ocr_rec_model")
-ocr_service.set_gpus("0")
 ocr_service.init_rec()
-ocr_service.prepare_server(workdir="workdir", port=9292, device="gpu", gpuid=0)
+if sys.argv[1] == 'gpu':
+    ocr_service.set_gpus("0")
+    ocr_service.prepare_server(workdir="workdir", port=9292, device="gpu", gpuid=0)
+elif sys.argv[1] == 'cpu':
+    ocr_service.prepare_server(workdir="workdir", port=9292, device="cpu")
 ocr_service.run_rpc_service()
 ocr_service.run_web_service()