2b 還是不 2b？這場 LLM 排程大賽可能改變你的 AI 應用方式

Hey everyone, I am generally interested in resource management and notably reducing the token cost for a given answer. So I just launched a Kaggle competition around a simple question: whether you should run a small model or not. I plan to add more model over time for better decision making. Here is the competition: https://www.kaggle.com/competitions/llm-scheduling-competition The idea: You get questions from the MMLU benchmark Instead of answering them, you decide: 2b → run a small mo