Fetching the paper…

FlexPrefill: A Context-Aware Sparse Attention Mechanism for Efficient Long-Sequence Inference · Around