hi! i used the average accuracy over the entire dataset made originally made by the llm