DeepSeek Sparse Attention: Boosting Long-Context Efficiency [pdf] | Dark Hacker News