arxiv Pushing The Limit of LLM Capacity for Text Classification