FlexPrefill: A Context-Aware Sparse Attention Mechanism for Efficient Long-Sequence Inference Paper • 2502.20766 • Published Feb 28 • 1