Abstract page for arXiv paper 2610.20641: AbridgeKV: Compressing Long Reasoning into the KV Cache of Learned Pause Tokens