fix

Author: shihaobai

lightllm/server/router/manager.py

@@ -113,12 +113,10 @@ async def wait_to_model_ready(self):
         ]
         self.rpc_event = multiprocessing.Event()
         self.rpc_finished_event = multiprocessing.Event()
-        
+
         assert (self.world_size % self.nnodes) == 0
         node_world_size = self.world_size // self.nnodes
-        for rank_id in range(
-            self.node_rank * node_world_size, (self.node_rank + 1) * node_world_size)
-        ):
+        for rank_id in range(self.node_rank * node_world_size, (self.node_rank + 1) * node_world_size):
             rpc_model = await start_model_process(
                 args=self.args,
                 rank=rank_id,

lightllm/server/router/model_infer/model_rpc.py

@@ -306,7 +306,7 @@ async def start_model_process(
     import lightllm.utils.rpyc_fix_utils as _
 
     # 单卡时不使用 rpc
-    if node_world_size == 1:
+    if node_world_size == 1 and args.nnodes == 1:
         return ModelRpcServer(
             args,
             rank,