implement grpc + local predictor

python/paddle_serving_app/local_predict.py

@@ -76,7 +76,7 @@ class Debugger(object):
         config.switch_use_feed_fetch_ops(False)
         self.predictor = create_paddle_predictor(config)
 
-    def predict(self, feed=None, fetch=None):
+    def predict(self, feed=None, fetch=None, batch=True):
         if feed is None or fetch is None:
             raise ValueError("You should specify feed and fetch for prediction")
         fetch_list = []
@@ -116,14 +116,25 @@ class Debugger(object):
 
         input_names = self.predictor.get_input_names()
         for name in input_names:
+            print(feed)
             if isinstance(feed[name], list):
                 feed[name] = np.array(feed[name]).reshape(self.feed_shapes_[
                     name])
             if self.feed_types_[name] == 0:
                 feed[name] = feed[name].astype("int64")
-            else:
+            elif self.feed_types_[name] == 1:
                 feed[name] = feed[name].astype("float32")
+            elif self.feed_types_[name] == 2:
+                feed[name] = feed[name].astype("int32")
+            else:
+                raise ValueError("local predictor receives wrong data type")
             input_tensor = self.predictor.get_input_tensor(name)
+            #TODO:set lods
+            if "{}.lod".format(name) in feed:
+                input_tensor.set_lod(feed["{}.lod".format(name)])
+            if batch == True:
+                input_tensor.copy_from_cpu(feed[name][np.newaxis,:])
+            else:
                 input_tensor.copy_from_cpu(feed[name])
         output_tensors = []
         output_names = self.predictor.get_output_names()

python/paddle_serving_server/__init__.py

@@ -453,14 +453,18 @@ class Server(object):
 
 class MultiLangServerServiceServicer(multi_lang_general_model_service_pb2_grpc.
                                      MultiLangGeneralModelServiceServicer):
-    def __init__(self, model_config_path, is_multi_model, endpoints):
-        self.is_multi_model_ = is_multi_model
+    def __init__(self, model_config_path, is_multi_model, endpoints, local_predictor=None):
         self.model_config_path_ = model_config_path
-        self.endpoints_ = endpoints
         with open(self.model_config_path_) as f:
             self.model_config_str_ = str(f.read())
         self._parse_model_config(self.model_config_str_)
+        self.is_multi_model_ = is_multi_model
+        if local_predictor == None:
+            self.local_predictor = None
+            self.endpoints_ = endpoints
             self._init_bclient(self.model_config_path_, self.endpoints_)
+        else:
+            self.local_predictor = local_predictor
 
     def _init_bclient(self, model_config_path, endpoints, timeout_ms=None):
         from paddle_serving_client import Client
@@ -585,8 +589,12 @@ class MultiLangServerServiceServicer(multi_lang_general_model_service_pb2_grpc.
     def Inference(self, request, context):
         feed_dict, fetch_names, is_python = self._unpack_inference_request(
             request)
+        if self.local_predictor == None:
             ret = self.bclient_.predict(
                     feed=feed_dict, fetch=fetch_names, need_variant_tag=True)
+        else:
+            ret = [self.local_predictor.predict(
+                    feed=feed_dict[0], fetch=fetch_names), "VariantTagNeeded"]
         return self._pack_inference_response(ret, fetch_names, is_python)
 
     def GetClientConfig(self, request, context):
@@ -596,8 +604,14 @@ class MultiLangServerServiceServicer(multi_lang_general_model_service_pb2_grpc.
 
 
 class MultiLangServer(object):
-    def __init__(self):
+    def __init__(self, use_local_predictor=False):
+        self.use_local_predictor = use_local_predictor
+        if use_local_predictor:
+            from paddle_serving_app.local_predict import Debugger
+            self.local_predictor_ = Debugger()
+        else:
             self.bserver_ = Server()
+            self.local_predictor_ = None
         self.worker_num_ = 4
         self.body_size_ = 64 * 1024 * 1024
         self.concurrency_ = 100000
@@ -620,6 +634,9 @@ class MultiLangServer(object):
                 "max_body_size is less than default value, will use default value in service."
             )
 
+    def set_use_local_predictor(self, mode):
+        self.use_local_predictor = True
+    
     def set_port(self, port):
         self.gport_ = port
 
@@ -645,6 +662,7 @@ class MultiLangServer(object):
         self.bserver_.use_mkl(flag)
 
     def load_model_config(self, server_config_paths, client_config_path=None):
+        if self.use_local_predictor == False:
             self.bserver_.load_model_config(server_config_paths)
             if client_config_path is None:
                 if isinstance(server_config_paths, dict):
@@ -655,6 +673,12 @@ class MultiLangServer(object):
                     client_config_path = '{}/serving_server_conf.prototxt'.format(
                         server_config_paths)
             self.bclient_config_path_ = client_config_path
+        else:
+            if isinstance(server_config_paths, dict):
+                raise ValueError("local predictor does not support model essemble")
+            client_config_path = '{}/serving_server_conf.prototxt'.format(server_config_paths)
+            self.local_predictor_.load_model_config(server_config_paths, profile=False) 
+            self.local_config_path_ = client_config_path
 
     def prepare_server(self,
                        workdir=None,
@@ -662,7 +686,8 @@ class MultiLangServer(object):
                        device="cpu",
                        cube_conf=None):
         if not self._port_is_available(port):
-            raise SystemExit("Prot {} is already used".format(port))
+            raise SystemExit("Port {} is already used".format(port))
+        if self.use_local_predictor == False:
             default_port = 12000
             self.port_list_ = []
             for i in range(1000):
@@ -687,20 +712,29 @@ class MultiLangServer(object):
         return result != 0
 
     def run_server(self):
+        if self.use_local_predictor is False:
+            print("brpc server process start")
             p_bserver = Process(
                 target=self._launch_brpc_service, args=(self.bserver_, ))
             p_bserver.start()
+
         options = [('grpc.max_send_message_length', self.body_size_),
                    ('grpc.max_receive_message_length', self.body_size_)]
         server = grpc.server(
             futures.ThreadPoolExecutor(max_workers=self.worker_num_),
             options=options,
             maximum_concurrent_rpcs=self.concurrency_)
+        if self.use_local_predictor is False:
             multi_lang_general_model_service_pb2_grpc.add_MultiLangGeneralModelServiceServicer_to_server(
                     MultiLangServerServiceServicer(
                         self.bclient_config_path_, self.is_multi_model_,
-                ["0.0.0.0:{}".format(self.port_list_[0])]), server)
+                        ["0.0.0.0:{}".format(self.port_list_[0])], None), server)
+        else:
+            multi_lang_general_model_service_pb2_grpc.add_MultiLangGeneralModelServiceServicer_to_server(
+                    MultiLangServerServiceServicer(
+                        self.local_config_path_, None, None, self.local_predictor_), server)
         server.add_insecure_port('[::]:{}'.format(self.gport_))
         server.start()
+        if self.use_local_predictor is False:
             p_bserver.join()
         server.wait_for_termination()

python/paddle_serving_server/serve.py

@@ -58,6 +58,11 @@ def parse_args():  # pylint: disable=doc-string-missing
         default=False,
         action="store_true",
         help="Use Multi-language-service")
+    parser.add_argument(
+        "--local_predict",
+        default=False,
+        action="store_true",
+        help="Use Local Predictor")
     return parser.parse_args()
 
 
@@ -73,6 +78,7 @@ def start_standard_model():  # pylint: disable=doc-string-missing
     max_body_size = args.max_body_size
     use_mkl = args.use_mkl
     use_multilang = args.use_multilang
+    local_predict=args.local_predict
 
     if model == "":
         print("You must specify your serving model")
@@ -91,17 +97,20 @@ def start_standard_model():  # pylint: disable=doc-string-missing
 
     server = None
     if use_multilang:
-        server = serving.MultiLangServer()
+        server = serving.MultiLangServer(local_predict)
     else:
+        if local_predict == True:
+            raise ValueError("local predict can only run with multilang")
         server = serving.Server()
+    if local_predict is False:
         server.set_op_sequence(op_seq_maker.get_op_sequence())
         server.set_num_threads(thread_num)
         server.set_memory_optimize(mem_optim)
         server.set_ir_optimize(ir_optim)
         server.use_mkl(use_mkl)
         server.set_max_body_size(max_body_size)
-    server.set_port(port)
 
+    server.set_port(port)
     server.load_model_config(model)
     server.prepare_server(workdir=workdir, port=port, device=device)
     server.run_server()