fix is_local bugs

README.md

@@ -6,7 +6,7 @@
 
 1. install dependencies
 
-- for mlx (macos arm):  `pip install -e ".[mlx]" && pip install -r requirements/mlx.txt`
+- for mlx (macos arm):  `pip install -U -e ".[mlx]" && pip install -e git+https://github.com/wnma3mz/mlx_clip.git#egg=mlx_clip`
 - for nvidia: `pip install -e ".[torch]"`
 
 2. run server

requirements/mlx.txt

@@ -1,4 +1,3 @@
 mlx==0.22.0
 mlx-lm==0.21.1
-mlx-vlm==0.1.12
--e git+https://github.com/wnma3mz/mlx_clip.git#egg=mlx_clip
+mlx-vlm==0.1.12

setup.py

@@ -9,7 +9,7 @@
 
 # 平台特定依赖
 with open(root_dir / "requirements" / "mlx.txt") as fid:
-    mlx_requires = [l.strip() for l in fid.readlines() if not l.startswith("-e")]
+    mlx_requires = [l.strip() for l in fid.readlines()]
 
 with open(root_dir / "requirements" / "torch.txt") as fid:
     torch_requires = [l.strip() for l in fid.readlines()]

tllm/entrypoints/api_server.py

@@ -151,7 +151,7 @@ async def update_model_url():
             for clients in host_list
         ]
         worker_rpc_manager.update_url(host_list)
-        master_url = args.hostname if is_local(args.hostname) else f"{args.hostname}:{args.grpc_port}"
+        master_url = args.hostname if args.is_local else f"{args.hostname}:{args.grpc_port}"
         await worker_rpc_manager.send_config(master_url, host_list)
         # 后台持续进行健康检查，如果有节点挂掉，需要重新分配
         await worker_rpc_manager.start_health_check()
@@ -185,7 +185,7 @@ async def init_model_func(
 
 async def init_app(engine: AsyncEngine, args):
     global app, openai_serving_chat, image_serving
-    logger.info("args: %s", args)
+    logger.info("Master Args: %s", args)
     if args.is_image:
         image_serving = ImageServing(engine, args)
     else:
@@ -225,7 +225,8 @@ async def run_server(args) -> None:
 
     uvicorn_kwargs = {"host": ["::", "0.0.0.0"], "port": args.http_port, "timeout_graceful_shutdown": 5}
 
-    if is_local(args.hostname):
+    args.is_local = is_local(args.hostname)
+    if args.is_local:
         if os.path.isfile(MASTER_SOCKET_PATH):
             os.remove(MASTER_SOCKET_PATH)
         if os.path.isfile(CLIENT_SOCKET_PATH):

tllm/entrypoints/utils.py

@@ -137,7 +137,7 @@ def start(self):
             return
         self.process = Process(target=self.run_server)
         self.process.start()
-        self.logger.info(f"Started gRPC process (PID: {self.process.pid})")
+        # self.logger.info(f"Worker gRPC process (PID: {self.process.pid})")
 
     def shutdown(self):
         """关闭 gRPC 服务器进程"""
@@ -148,7 +148,7 @@ def shutdown(self):
                 self.process.join(timeout=5)
                 if self.process.is_alive():
                     self.process.kill()
-            self.logger.info("gRPC process stopped")
+            self.logger.info("Worker gRPC process stopped")
 
 
 async def serve_http(app: FastAPI, grpc_process, engine, master_server, **uvicorn_kwargs: Dict):

tllm/grpc/master_service/master_server.py

@@ -30,12 +30,13 @@ async def stop(self):
                 await self.server.stop(grace=5)
                 await self.server.wait_for_termination()
             except (Exception, asyncio.CancelledError) as e:
-                print("master handler error", str(e))
+                self.logger.info("master handler error", str(e))
 
     async def Forward(
         self, request: schemas_pb2.ForwardRequest, context: grpc.ServicerContext
     ) -> schemas_pb2.ForwardResponse:
         """处理从最后一个节点返回的结果"""
+        self.logger.info("master handler request")
         request_id = "-".join(x for x in list(request.uuid_list))
 
         try:

tllm/grpc/worker_service/worker_server.py

@@ -41,7 +41,7 @@ async def start(self, ip_addr_list: List[str], port: int = 50051):
         schemas_pb2_grpc.add_RPCServiceServicer_to_server(self, self.server)
         self.server.add_insecure_port(f"[::]:{port}")
         self.server.add_insecure_port(f"unix://{CLIENT_SOCKET_PATH}")
-        self.logger.info(f"Starting gRPC server on [::]:{port}")
+        self.logger.info(f"Starting Worker gRPC server on [::]:{port}")
         await self.server.start()
 
         self.http_client.is_running = True
@@ -174,7 +174,7 @@ async def run(args):
 
     client_id = f"test-{str(uuid.uuid4())[:8]}-{comm.rank}"
     rpc_servicer = WorkerServer(comm, logger, args.master_addr, client_id)
-    logger.info("args: %s", args)
+    logger.info("Worker Args: %s", args)
     try:
         await rpc_servicer.start(ip_addr_list, args.grpc_port)
     except Exception as e: