run arm cpu bugfix use webservice;

brpc reduce BTHREAD_MIN_CONCURRENCY from 4 to 2

cmake/external/brpc.cmake

@@ -36,7 +36,7 @@ INCLUDE_DIRECTORIES(${BRPC_INCLUDE_DIR})
 set(prefix_path "${THIRD_PARTY_PATH}/install/gflags|${THIRD_PARTY_PATH}/install/leveldb|${THIRD_PARTY_PATH}/install/snappy|${THIRD_PARTY_PATH}/install/gtest|${THIRD_PARTY_PATH}/install/protobuf|${THIRD_PARTY_PATH}/install/zlib|${THIRD_PARTY_PATH}/install/glog")
 
 if(WITH_LITE)
-    set(BRPC_REPO "https://github.com/apache/incubator-brpc.git")
+    set(BRPC_REPO "https://github.com/zhangjun/incubator-brpc.git")
     set(BRPC_TAG "master")
 else()
     set(BRPC_REPO "https://github.com/wangjiawei04/brpc")

paddle_inference/inferencer-fluid-arm/src/fluid_arm_engine.cpp

@@ -19,6 +19,8 @@ namespace baidu {
 namespace paddle_serving {
 namespace fluid_arm {
 
+DEFINE_int32(gpuid, 0, "device id to use");
+
 REGIST_FACTORY_OBJECT_IMPL_WITH_NAME(
     ::baidu::paddle_serving::predictor::FluidInferEngine<FluidArmAnalysisCore>,
     ::baidu::paddle_serving::predictor::InferEngine,

python/paddle_serving_server_gpu/__init__.py

@@ -423,10 +423,12 @@ class Server(object):
         for line in version_file.readlines():
             if re.match("cuda_version", line):
                 cuda_version = line.split("\"")[1]
-                if cuda_version != "trt":
-                    device_version = "serving-gpu-cuda" + cuda_version + "-"
-                else:
+                if cuda_version == "trt":
                     device_version = "serving-gpu-" + cuda_version + "-"
+                elif cuda_version == "arm":
+                    device_version = "serving-" + cuda_version + "-"
+                elif:
+                    device_version = "serving-gpu-cuda" + cuda_version + "-"
 
         folder_name = device_version + serving_server_version
         tar_name = folder_name + ".tar.gz"

python/paddle_serving_server_gpu/serve.py

@@ -140,7 +140,8 @@ if __name__ == "__main__":
         if len(gpu_ids) > 0:
             web_service.set_gpus(gpu_ids)
         web_service.prepare_server(
-            workdir=args.workdir, port=args.port, device=args.device)
+            workdir=args.workdir, port=args.port, device=args.device,
+            use_lite=args.use_lite, use_xpu=args.use_xpu, ir_optim=args.ir_optim)
         web_service.run_rpc_service()
 
         app_instance = Flask(__name__)

python/paddle_serving_server_gpu/web_service.py

@@ -83,10 +83,15 @@ class WebService(object):
                             gpuid=0,
                             thread_num=2,
                             mem_optim=True,
+                            use_lite=False,
+                            use_xpu=False,
                             ir_optim=False):
         device = "gpu"
         if gpuid == -1:
-            device = "cpu"
+            if use_lite:
+                device = "arm"
+            else:
+                device = "cpu"
         op_maker = serving.OpMaker()
         read_op = op_maker.create('general_reader')
         general_infer_op = op_maker.create('general_infer')
@@ -103,6 +108,11 @@ class WebService(object):
         server.set_memory_optimize(mem_optim)
         server.set_ir_optimize(ir_optim)
 
+        if use_lite:
+            server.set_lite()
+        if use_xpu:
+            server.set_xpu()
+
         server.load_model_config(self.model_config)
         if gpuid >= 0:
             server.set_gpuid(gpuid)
@@ -125,9 +135,11 @@ class WebService(object):
                        workdir="",
                        port=9393,
                        device="gpu",
+                       use_lite=False,
+                       use_xpu=False,
+                       ir_optim=False,
                        gpuid=0,
-                       mem_optim=True,
-                       ir_optim=False):
+                       mem_optim=True):
         print("This API will be deprecated later. Please do not use it")
         self.workdir = workdir
         self.port = port
@@ -150,6 +162,8 @@ class WebService(object):
                     -1,
                     thread_num=2,
                     mem_optim=mem_optim,
+                    use_lite=use_lite,
+                    use_xpu=use_xpu,
                     ir_optim=ir_optim))
         else:
             for i, gpuid in enumerate(self.gpus):
@@ -160,6 +174,8 @@ class WebService(object):
                         gpuid,
                         thread_num=2,
                         mem_optim=mem_optim,
+                        use_lite=use_lite,
+                        use_xpu=use_xpu,
                         ir_optim=ir_optim))
 
     def _launch_web_service(self):

python/setup.py.app.in

@@ -32,7 +32,7 @@ if '${PACK}' == 'ON':
 
 
 REQUIRED_PACKAGES = [
-    'six >= 1.10.0', 'sentencepiece<=0.1.92', 'opencv-python<=4.2.0.32', 'pillow',
+    'six >= 1.10.0', 'sentencepiece', 'opencv-python', 'pillow',
     'pyclipper'
 ]