fix export

src/llmtuner/train/tuner.py

@@ -60,8 +60,9 @@ def export_model(args: Optional[Dict[str, Any]] = None):
 
     if getattr(model, "quantization_method", None) is None:  # cannot convert dtype of a quantized model
         output_dtype = getattr(model.config, "torch_dtype", torch.float16)
-        model = model.to(output_dtype)
         setattr(model.config, "torch_dtype", output_dtype)
+        for param in model.parameters():
+            param.data = param.data.to(output_dtype)        
 
     model.save_pretrained(
         save_directory=model_args.export_dir,