CritiPrefill: A Segment-wise Criticality-based Approach for Prefilling Acceleration in LLMs.
Junlin Lv, Yuan Feng, Xike Xie, Xin Jia, Qirong Peng, Guiming Xie
Browse the full ICASSP paper archive.
Junlin Lv, Yuan Feng, Xike Xie, Xin Jia, Qirong Peng, Guiming Xie
Browse the full ICASSP paper archive.