Fetching the paper…

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