Our paper TweakLLM is now on arXiv (2507.23674) — adaptive caching that cuts LLM compute costs by up to 68%.