KOALA: Enhancing Speculative Decoding for LLM via Multi-Layer Draft Heads with Adversarial Learning.

Kaiqi Zhang, Jing Zhao, Rui Chen

Browse the full CSCWD paper archive.