Robust and Efficient Early Exit for Large Language Models: Mitigating KV Cache Loss and Enhancing Exit Stability.

Long Meng, Ruiqing Zhang, Weiqiao Shan

Browse the full ISNN paper archive.