Update Neural Solution based on v2.4 release test (#1410)

Signed-off-by: Kaihui-intel <kaihui.tang@intel.com>

neural_solution/backend/scheduler.py

@@ -271,7 +271,7 @@ def launch_task(self, task: Task, resource):
         )
         task_status = self.check_task_status(log_path)
         self.task_db.update_task_status(task.task_id, task_status)
-        self.q_model_path = get_q_model_path(log_path=log_path) if task_status == "done" else None
+        self.q_model_path = get_q_model_path(log_path=log_path, task_id=task.task_id) if task_status == "done" else None
         self.report_result(task.task_id, log_path, task_runtime)
 
     def dispatch_task(self, task, resource):

neural_solution/backend/utils/utility.py

@@ -226,11 +226,12 @@ def create_dir(path):
         os.makedirs(os.path.dirname(path))
 
 
-def get_q_model_path(log_path):
+def get_q_model_path(log_path, task_id):
     """Get the quantized model path from task log.
 
     Args:
         log_path (str): log path for task
+        task_id: the id of task
 
     Returns:
         str: quantized model path
@@ -241,5 +242,8 @@ def get_q_model_path(log_path):
         match = re.search(r"(Save quantized model to|Save config file and weights of quantized model to) (.+?)\.", line)
         if match:
             q_model_path = match.group(2)
+            match_task_id = re.search(r"(.+/task_workspace/{}/[^/]+)".format(task_id), q_model_path)
+            if match_task_id:
+                q_model_path = match_task_id.group()
             return q_model_path
     return "quantized model path not found"

neural_solution/examples/hf_models/task_request.json

@@ -5,6 +5,6 @@
         "--model_name_or_path bert-base-cased --task_name mrpc --do_eval --output_dir result"
     ],
     "approach": "static",
-    "requirements": [],
+    "requirements": ["datasets", "transformers=4.21.0", "torch"],
     "workers": 1
-}
+}

neural_solution/examples/hf_models_grpc/README.md

@@ -80,6 +80,7 @@ optional arguments:
 - Step 2: Submit the task request to service, and it will return the submit status and task id for future use.
 
 ```shell
+[user@server hf_models_grpc]$ cd path/to/neural_solution/examples/hf_models_grpc
 [user@server hf_models_grpc]$ python client.py submit --request="test_task_request.json"
 
 # response if submit successfully

neural_solution/examples/hf_models_grpc/task_request.json

@@ -5,6 +5,6 @@
         "--model_name_or_path bert-base-cased --task_name mrpc --do_eval --output_dir result"
     ],
     "approach": "static",
-    "requirements": [],
+    "requirements": ["datasets", "transformers=4.21.0", "torch"],
     "workers": 1
-}
+}

neural_solution/frontend/gRPC/client.py

@@ -126,9 +126,9 @@ def run_query_task_status(args):
     query_action_parser.set_defaults(func=run_query_task_result)
     query_action_parser.add_argument("--task_id", type=str, default=None, help="Query task by task id.")
 
-    parser.add_argument("--grpc_api_port", type=str, default=None, help="grpc server port.")
-    parser.add_argument("--result_monitor_port", type=str, default=None, help="result monitor port.")
-    parser.add_argument("--task_monitor_port", type=str, default=None, help="task monitor port.")
+    parser.add_argument("--grpc_api_port", type=str, default="8001", help="grpc server port.")
+    parser.add_argument("--result_monitor_port", type=str, default="2222", help="result monitor port.")
+    parser.add_argument("--task_monitor_port", type=str, default="3333", help="task monitor port.")
 
     args = parser.parse_args()
     config.grpc_api_port = args.grpc_api_port

neural_solution/test/backend/utils/test_utility.py

@@ -138,13 +138,19 @@ def test_create_dir(self):
     @patch("builtins.open", mock_open(read_data="Save quantized model to /path/to/model."))
     def test_get_q_model_path_success(self):
         log_path = "fake_log_path"
-        q_model_path = get_q_model_path(log_path)
+        q_model_path = get_q_model_path(log_path, "task_id")
         self.assertEqual(q_model_path, "/path/to/model")
 
+    @patch("builtins.open", mock_open(read_data="Save quantized model to /path/to/task_workspace/task_id/model/1.pb."))
+    def test_get_q_model_path_success_task_id(self):
+        log_path = "fake_log_path"
+        q_model_path = get_q_model_path(log_path, "task_id")
+        self.assertEqual(q_model_path, "/path/to/task_workspace/task_id/model")
+
     @patch("builtins.open", mock_open(read_data="No quantized model saved."))
     def test_get_q_model_path_failure(self):
         log_path = "fake_log_path"
-        q_model_path = get_q_model_path(log_path)
+        q_model_path = get_q_model_path(log_path, "task_id")
         self.assertEqual(q_model_path, "quantized model path not found")