diff --git a/xtuner/tools/model_converters/pth_to_hf.py b/xtuner/tools/model_converters/pth_to_hf.py
index 33a18e0df..2a4b28883 100644
--- a/xtuner/tools/model_converters/pth_to_hf.py
+++ b/xtuner/tools/model_converters/pth_to_hf.py
@@ -83,6 +83,9 @@ def main():
     if 'Reward' in model_name:
         use_meta_init = False
         cfg.model.llm.pop('quantization_config', None)
+    if hasattr(cfg.model.llm, 'quantization_config'):
+        # Can not build a qlora model on meta device
+        use_meta_init = False
 
     if use_meta_init:
         try: