use_exllama=false?

app.py

@@ -16,7 +16,8 @@ tokenizer = AutoTokenizer.from_pretrained(model_id)
 quantize_config = BaseQuantizeConfig(
         bits=4,
         group_size=128,
-        desc_act=False
+        desc_act=False,
+        use_exllama=False
     )
 model = AutoGPTQForCausalLM.from_quantized(
         model_id,