Speculative Decoding via Early-exiting for Faster LLM Inference with Thompson Sampling Control Mechanism.

Jiahao Liu, Qifan Wang, Jingang Wang, Xunliang Cai

Browse the full ACL paper archive.