Native Sparse Attention: Hardware-Aligned and Natively Trainablearxiv.org2 points·teepo··0 commentsOpen articleSaveView on HN