3 ms·HW-Aligned Sparse Attention Architecture for Efficient Long-Context Modeling2 points by PaulHoule 2y ago