diff --git a/autograd_4bit.py b/autograd_4bit.py
index 476da12..5a47921 100644
--- a/autograd_4bit.py
+++ b/autograd_4bit.py
@@ -248,7 +248,7 @@ def load_llama_model_4bit_low_ram_and_offload(config_path, model_path, lora_path
 
     if lora_path is not None:
         from peft import PeftModel
-        from peft.tuners.lora import Linear4bitLt
+        from monkeypatch.peft_tuners_lora_monkey_patch import Linear4bitLt
         model = PeftModel.from_pretrained(model, lora_path, device_map={'': 'cpu'}, torch_dtype=torch.float32)
         print(Style.BRIGHT + Fore.GREEN + '{} Lora Applied.'.format(lora_path))
 
diff --git a/text-generation-webui/custom_monkey_patch.py b/text-generation-webui/custom_monkey_patch.py
index c59132e..28bb53e 100644
--- a/text-generation-webui/custom_monkey_patch.py
+++ b/text-generation-webui/custom_monkey_patch.py
@@ -3,7 +3,8 @@ import torch
 import autograd_4bit
 from autograd_4bit import load_llama_model_4bit_low_ram, Autograd4bitQuantLinear
 from peft import PeftModel
-from peft.tuners.lora import Linear4bitLt
+from monkeypatch.peft_tuners_lora_monkey_patch import replace_peft_model_with_gptq_lora_model, Linear4bitLt
+replace_peft_model_with_gptq_lora_model()
 
 patch_encode_func = False