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

Open in new window