diff --git a/modules/lora/lora_apply.py b/modules/lora/lora_apply.py
index 0c3ff6ef3..a4aa18c70 100644
--- a/modules/lora/lora_apply.py
+++ b/modules/lora/lora_apply.py
@@ -275,7 +275,7 @@ def network_apply_weights(self: torch.nn.Conv2d | torch.nn.Linear | torch.nn.Gro
                     self.svd_down = torch.nn.Parameter(self.sdnq_svd_down_backup.to(device), requires_grad=False)
                 else:
                     self.svd_up, self.svd_down = None, None
-                del self.sdnq_dequantizer_backup, self.sdnq_scale_backup, self.sdnq_zero_point_backup, self.sdnq_svd_up_backup, self.sdnq_svd_down_backup
+                # del self.sdnq_dequantizer_backup, self.sdnq_scale_backup, self.sdnq_zero_point_backup, self.sdnq_svd_up_backup, self.sdnq_svd_down_backup
 
     if bias_backup is not None:
         self.bias = None