[Researchers propose Dual Chunk Attention (DCA) to improve Large Language Models (LLMs) processing long sequences, offering comparable performance gains.]