Less Is More: Training-Free Sparse Attention with Global Locality for Efficient Reasoning

Open in new window