How to avoid LLM prompt token limit

This is a prompt for normalize the results, and need help as token limit is being hit each time when this prompt is being sent to LLM API. WHat are the options other than chunking? def _load_kimi_prompt() -> str: return """ Normalize ACORD 140 "Subject of Insurance" rows. The text you return is matched downstream by literal keyword checks. Input is {"rows":[{"id","text"}]}. Return exactly one output row per input row. The text is OCR of scanned forms. Expect and silently repair: ALL CAPS; ...