AbridgeKV: Compressing Long Reasoning into the KV Cache of Learned Pause Tokensgithub.com1 point·matt_d··0 commentsOpen articleSaveView on HN