fix bug

finetune.py

@@ -172,6 +172,9 @@ if not ft_config.skip:
     else:
         trainer.train()
 
+    # Restore old model state dict
+    model.state_dict = old_state_dict
+
     print('Train completed.')
 
 # Save Model