diff --git a/python/CMakeLists.txt b/python/CMakeLists.txt
index f3762df4616e7e971772b2955954af946132329f..6fb78421fbb1241d49ab512f7f2c73df96b563a1 100644
--- a/python/CMakeLists.txt
+++ b/python/CMakeLists.txt
@@ -91,15 +91,27 @@ if (SERVER)
             DEPENDS ${SERVING_SERVER_CORE} server_config_py_proto ${PY_FILES})
         add_custom_target(paddle_python ALL DEPENDS ${PADDLE_SERVING_BINARY_DIR}/.timestamp)
     elseif(WITH_LITE)
-        add_custom_command(
-            OUTPUT ${PADDLE_SERVING_BINARY_DIR}/.timestamp
-            COMMAND cp -r
-            ${CMAKE_CURRENT_SOURCE_DIR}/paddle_serving_server_gpu/ ${PADDLE_SERVING_BINARY_DIR}/python/
-            COMMAND env ${py_env} ${PYTHON_EXECUTABLE} gen_version.py
-            "server_gpu" arm 
-            COMMAND env ${py_env} ${PYTHON_EXECUTABLE} setup.py bdist_wheel
-            DEPENDS ${SERVING_SERVER_CORE} server_config_py_proto ${PY_FILES})
-        add_custom_target(paddle_python ALL DEPENDS ${PADDLE_SERVING_BINARY_DIR}/.timestamp)
+        if(WITH_XPU)
+            add_custom_command(
+                OUTPUT ${PADDLE_SERVING_BINARY_DIR}/.timestamp
+                COMMAND cp -r
+                ${CMAKE_CURRENT_SOURCE_DIR}/paddle_serving_server_gpu/ ${PADDLE_SERVING_BINARY_DIR}/python/
+                COMMAND env ${py_env} ${PYTHON_EXECUTABLE} gen_version.py
+                "server_gpu" arm-xpu 
+                COMMAND env ${py_env} ${PYTHON_EXECUTABLE} setup.py bdist_wheel
+                DEPENDS ${SERVING_SERVER_CORE} server_config_py_proto ${PY_FILES})
+            add_custom_target(paddle_python ALL DEPENDS ${PADDLE_SERVING_BINARY_DIR}/.timestamp)
+        else()
+            add_custom_command(
+                OUTPUT ${PADDLE_SERVING_BINARY_DIR}/.timestamp
+                COMMAND cp -r
+                ${CMAKE_CURRENT_SOURCE_DIR}/paddle_serving_server_gpu/ ${PADDLE_SERVING_BINARY_DIR}/python/
+                COMMAND env ${py_env} ${PYTHON_EXECUTABLE} gen_version.py
+                "server_gpu" arm 
+                COMMAND env ${py_env} ${PYTHON_EXECUTABLE} setup.py bdist_wheel
+                DEPENDS ${SERVING_SERVER_CORE} server_config_py_proto ${PY_FILES})
+            add_custom_target(paddle_python ALL DEPENDS ${PADDLE_SERVING_BINARY_DIR}/.timestamp)
+        endif()
     else()
         add_custom_command(
             OUTPUT ${PADDLE_SERVING_BINARY_DIR}/.timestamp
diff --git a/python/paddle_serving_server_gpu/__init__.py b/python/paddle_serving_server_gpu/__init__.py
index 13f6a61c600995be95b051d3b2691ae68e5e788e..0182f1cf8907b42d3e0c2c79c19dc6779452120a 100644
--- a/python/paddle_serving_server_gpu/__init__.py
+++ b/python/paddle_serving_server_gpu/__init__.py
@@ -311,8 +311,9 @@ class Server(object):
             engine.static_optimization = False
             engine.force_update_static_cache = False
             engine.use_trt = self.use_trt
-            engine.use_lite = self.use_lite
-            engine.use_xpu = self.use_xpu
+            if device == "arm":
+                engine.use_lite = self.use_lite
+                engine.use_xpu = self.use_xpu
 
 
 
@@ -425,7 +426,7 @@ class Server(object):
                 cuda_version = line.split("\"")[1]
                 if cuda_version == "trt":
                     device_version = "serving-gpu-" + cuda_version + "-"
-                elif cuda_version == "arm":
+                elif cuda_version == "arm" or cuda_version == "arm-xpu":
                     device_version = "serving-" + cuda_version + "-"
                 else:
                     device_version = "serving-gpu-cuda" + cuda_version + "-"