Oc-e-Non-v1: A language model that gets to the f*cking point
Richard I of England was called "Oc-e-Non" ( Yes and No in Occitan) by some of contemporaries, supposedly because he used few words. Wouldn't it be nice if LLMs followed his shining example? Oc-e-Non-v1 is a tersified version of Qwen3-4B-Instruct-2507. It reduces Qwen's mean response length by 50%.