A way to cache llm response Normal kv cache is not applicable for llm response
https://youtu.be/1RxOYLa69Vw