MInference 1.0: Accelerating Pre-filling for Long-Context LLMs via Dynamic Sparse Attention

Jiang, Huiqiang, Li, Yucheng, Zhang, Chengruidong · arXiv · 2024