init

vector_app.py

@@ -8,8 +8,14 @@ from routes import server_routes
 
 app = Quart(__name__)
 
-# llm
-llm = None
+MODEL_PATH = LOCAL_MODEL_CONFIG[DEFAULT_MODEL]
+
+llm = LLM(
+    model=MODEL_PATH,
+    dtype="float16",   # 节省显存
+    trust_remote_code=True
+)
+print(f"{MODEL_PATH} 模型加载完成！")
 
 # 连接向量数据库
 # connections.connect("default", host="milvus", port="19530")
@@ -20,16 +26,3 @@ connections = None
 app_route = server_routes(llm, connections)
 app.register_blueprint(app_route)
 
-@app.before_serving
-async def load_model():
-    """在服务启动前加载模型"""
-    global llm
-    MODEL_PATH = LOCAL_MODEL_CONFIG[DEFAULT_MODEL]
-    if llm is None:
-        llm = LLM(
-            model=MODEL_PATH,
-            dtype="float16",   # 节省显存
-            trust_remote_code=True
-        )
-        print(f"{MODEL_PATH} 模型加载完成！")
-