Update app.py

app.py

@@ -56,7 +56,7 @@ model = transformers.AutoModelForCausalLM.from_pretrained(
             model_id,
             trust_remote_code=True,
             config=model_config,
-            quantization_config=bnb_config,
+            # quantization_config=bnb_config,
             token=HF_TOKEN,
             # use_auth_token=hf_auth
         )