3 ms·Π-Attention: Periodic Sparse Transformers for Efficient Long-Context Modeling1 points by PaulHoule 1y ago