修复一个data_parallel情况下的gpu不从0开始导致的bug

fastNLP/core/utils.py

@@ -409,7 +409,7 @@ def _move_model_to_device(model, device):
 
     if device is None:
         if isinstance(model, torch.nn.DataParallel):
-            model.cuda()
+            model.cuda(model.device_ids[0])
         return model
     else:
         if not torch.cuda.is_available() and ((isinstance(device, str) and device!='cpu') or