DuoAttention-Slashes memory and latency for LLMs without sacrificing performancegithub.com2 points·dsr12··0 commentsOpen articleSaveView on HN