Token Sparse Attention: Efficient Long-Context Inference with Interleaved Token Selection Paper • 2602.03216 • Published Feb 3 • 12