Less Is More: Training-Free Sparse Attention with Global Locality for Efficient Reasoning Paper • 2508.07101 • Published Aug 9 • 13 • 2
