diff --git a/jobs/ExtensionJob.py b/jobs/ExtensionJob.py
index cad870cf..5ca4ff73 100644
--- a/jobs/ExtensionJob.py
+++ b/jobs/ExtensionJob.py
@@ -21,33 +21,21 @@ def __init__(self, config: OrderedDict):
                 # 多 GPU 情况
                 self.distributed = True
                 self.local_rank = int(os.getenv('LOCAL_RANK', 0))
-                self.device = torch.device(f'cuda:{self.local_rank}')
+                if self.local_rank < len(devices):
+                    self.device = torch.device(f'cuda:{devices[self.local_rank].strip()}')  # 确保设备字符串正确
+                else:
+                    raise RuntimeError(f"Invalid LOCAL_RANK {self.local_rank}, exceeding available devices {devices}")
                 self.world_size = len(devices)
                 dist.init_process_group(backend='nccl', rank=self.local_rank, world_size=self.world_size)
             else:
                 # 单 GPU 情况
                 self.distributed = False
-                self.device = torch.device(self.device_config)
+                self.device = torch.device(self.device_config.strip())  # 单 GPU 情况，去除空格
         else:
             # 默认 CPU 情况
             self.distributed = False
             self.device = torch.device('cpu')
 
-        # 从配置文件加载模型路径或配置参数
-        model_path = self.get_conf('model_path', None)  # 假设配置文件中定义了 model_path
-        if model_path:
-            # 加载预训练模型
-            self.model = torch.load(model_path, map_location=self.device)
-        else:
-            raise ValueError("模型路径未在配置文件中指定，请检查 'model_path' 配置项。")
-
-        # 如果是分布式训练模式，使用 DDP 包装模型
-        if self.distributed:
-            self.model = DDP(self.model, device_ids=[self.local_rank], output_device=self.local_rank)
-
-        # 将模型移动到相应设备
-        self.model = self.model.to(self.device)
-
         # 加载扩展的进程
         self.process_dict = get_all_extensions_process_dict()
         self.load_processes(self.process_dict)