optimize memory

README.md

@@ -124,7 +124,7 @@ python -m paddle_serving_server.serve --model uci_housing_model --thread 10 --po
 | `port` | int | `9292` | Exposed port of current service to users|
 | `name` | str | `""` | Service name, can be used to generate HTTP request url |
 | `model` | str | `""` | Path of paddle model directory to be served |
-| `mem_optim` | - | - | Enable memory / graphic memory optimization |
+| `mem_optim_off` | - | - | Disable memory / graphic memory optimization |
 | `ir_optim` | - | - | Enable analysis and optimization of calculation graph |
 | `use_mkl` (Only for cpu version) | - | - | Run inference with MKL |
 

README_CN.md

@@ -120,7 +120,7 @@ python -m paddle_serving_server.serve --model uci_housing_model --thread 10 --po
 | `port` | int | `9292` | Exposed port of current service to users|
 | `name` | str | `""` | Service name, can be used to generate HTTP request url |
 | `model` | str | `""` | Path of paddle model directory to be served |
-| `mem_optim` | - | - | Enable memory optimization |
+| `mem_optim_off` | - | - | Disable memory optimization |
 | `ir_optim` | - | - | Enable analysis and optimization of calculation graph |
 | `use_mkl` (Only for cpu version) | - | - | Run inference with MKL |
 

cmake/external/brpc.cmake

@@ -40,8 +40,8 @@ ExternalProject_Add(
     extern_brpc
     ${EXTERNAL_PROJECT_LOG_ARGS}
     # TODO(gongwb): change to de newst repo when they changed.
-    GIT_REPOSITORY  "https://github.com/gongweibao/brpc"
-    GIT_TAG         "e9b67ec1b7458f2af5fae76451afe1e27e01b4b4"
+    GIT_REPOSITORY  "https://github.com/wangjiawei04/brpc"
+    GIT_TAG         "6d79e0b17f25107c35b705ea58d888083f59ff47"
     PREFIX          ${BRPC_SOURCES_DIR}
     UPDATE_COMMAND  ""
     CMAKE_ARGS      -DCMAKE_CXX_COMPILER=${CMAKE_CXX_COMPILER}

doc/FAQ.md

@@ -13,6 +13,3 @@
   client.set_rpc_timeout_ms(100000)
   client.connect(["127.0.0.1:9393"])
    ```
-
-- Q: 执行GPU预测时出现显存不足的问题应该怎么办？
-  A: 请使用--thread 4参数限制server端的线程数为4，并使用--mem_optim参数开启显存优化选项。

python/paddle_serving_server/serve.py

@@ -40,7 +40,7 @@ def parse_args():  # pylint: disable=doc-string-missing
     parser.add_argument(
         "--device", type=str, default="cpu", help="Type of device")
     parser.add_argument(
-        "--mem_optim",
+        "--mem_optim_off",
         default=False,
         action="store_true",
         help="Memory optimize")
@@ -68,7 +68,7 @@ def start_standard_model():  # pylint: disable=doc-string-missing
     port = args.port
     workdir = args.workdir
     device = args.device
-    mem_optim = args.mem_optim
+    mem_optim = args.mem_optim_off is False
     ir_optim = args.ir_optim
     max_body_size = args.max_body_size
     use_mkl = args.use_mkl

python/paddle_serving_server/web_service.py

@@ -41,6 +41,8 @@ class WebService(object):
         server = Server()
         server.set_op_sequence(op_seq_maker.get_op_sequence())
         server.set_num_threads(16)
+        server.set_memory_optimize(self.mem_optim)
+        server.set_ir_optimize(self.ir_optim)
         server.load_model_config(self.model_config)
         server.prepare_server(
             workdir=self.workdir, port=self.port_list[0], device=self.device)
@@ -55,12 +57,19 @@ class WebService(object):
         else:
             return False
 
-    def prepare_server(self, workdir="", port=9393, device="cpu"):
+    def prepare_server(self,
+                       workdir="",
+                       port=9393,
+                       device="cpu",
+                       mem_optim=True,
+                       ir_optim=False):
         self.workdir = workdir
         self.port = port
         self.device = device
         default_port = 12000
         self.port_list = []
+        self.mem_optim = mem_optim
+        self.ir_optim = ir_optim
         for i in range(1000):
             if self.port_is_available(default_port + i):
                 self.port_list.append(default_port + i)

python/paddle_serving_server_gpu/__init__.py

@@ -41,7 +41,7 @@ from concurrent import futures
 def serve_args():
     parser = argparse.ArgumentParser("serve")
     parser.add_argument(
-        "--thread", type=int, default=4, help="Concurrency of server")
+        "--thread", type=int, default=2, help="Concurrency of server")
     parser.add_argument(
         "--model", type=str, default="", help="Model for serving")
     parser.add_argument(
@@ -57,7 +57,7 @@ def serve_args():
     parser.add_argument(
         "--name", type=str, default="None", help="Default service name")
     parser.add_argument(
-        "--mem_optim",
+        "--mem_optim_off",
         default=False,
         action="store_true",
         help="Memory optimize")
@@ -187,7 +187,7 @@ class Server(object):
         self.cube_config_fn = "cube.conf"
         self.workdir = ""
         self.max_concurrency = 0
-        self.num_threads = 4
+        self.num_threads = 2
         self.port = 8080
         self.reload_interval_s = 10
         self.max_body_size = 64 * 1024 * 1024

python/paddle_serving_server_gpu/serve.py

@@ -34,7 +34,7 @@ def start_gpu_card_model(index, gpuid, args):  # pylint: disable=doc-string-miss
         port = args.port + index
     thread_num = args.thread
     model = args.model
-    mem_optim = args.mem_optim
+    mem_optim = args.mem_optim_off is False
     ir_optim = args.ir_optim
     max_body_size = args.max_body_size
     use_multilang = args.use_multilang

python/paddle_serving_server_gpu/web_service.py

@@ -41,7 +41,9 @@ class WebService(object):
                             workdir="conf",
                             port=9292,
                             gpuid=0,
-                            thread_num=10):
+                            thread_num=2,
+                            mem_optim=True,
+                            ir_optim=False):
         device = "gpu"
         if gpuid == -1:
             device = "cpu"
@@ -58,6 +60,8 @@ class WebService(object):
         server = Server()
         server.set_op_sequence(op_seq_maker.get_op_sequence())
         server.set_num_threads(thread_num)
+        server.set_mem_optim(mem_optim)
+        server.set_ir_optim(ir_optim)
 
         server.load_model_config(self.model_config)
         if gpuid >= 0:
@@ -77,7 +81,13 @@ class WebService(object):
         else:
             return False
 
-    def prepare_server(self, workdir="", port=9393, device="gpu", gpuid=0):
+    def prepare_server(self,
+                       workdir="",
+                       port=9393,
+                       device="gpu",
+                       gpuid=0,
+                       mem_optim=True,
+                       ir_optim=False):
         self.workdir = workdir
         self.port = port
         self.device = device
@@ -94,7 +104,12 @@ class WebService(object):
             # init cpu service
             self.rpc_service_list.append(
                 self.default_rpc_service(
-                    self.workdir, self.port_list[0], -1, thread_num=10))
+                    self.workdir,
+                    self.port_list[0],
+                    -1,
+                    thread_num=2,
+                    mem_optim,
+                    ir_optim))
         else:
             for i, gpuid in enumerate(self.gpus):
                 self.rpc_service_list.append(
@@ -102,7 +117,9 @@ class WebService(object):
                         "{}_{}".format(self.workdir, i),
                         self.port_list[i],
                         gpuid,
-                        thread_num=10))
+                        thread_num=2,
+                        mem_optim,
+                        ir_optim))
 
     def _launch_web_service(self):
         gpu_num = len(self.gpus)