Update app.py

app.py

@@ -78,7 +78,10 @@ def inference(model_list, question, API_KEY, cot, hf_token=HF_TOKEN):
         raise gr.Error("Please choose just '3' models! Neither more nor less!")
 
     for i in range(len(model_list)):
-        model_list[i] = model_list[i][:-1].lower()
+        if "llama" in model_list[I].lower():
+            model_list[i] = model_list[i][:-2].lower()
+        else:
+            model_list[i] = model_list[i][:-1].lower()
 
     model_response = Inference(model_list, question, API_KEY, cot, hf_token)