作者位次 #3/4ACL 2026Poster· Natural Language Processing
Jakiro: Boosting Speculative Decoding via Decoupled M o E
Haiduo Huang · Fuwei Yang · Zhenhua Liu · pengju ren
摘要
Speculative decoding has emerged as a promising technique to accelerate large language model inference by employing a smaller draft model to predict multiple tokens, which are then verified in parallel by the larger target model. However, existing approaches face a fundamental limitation: candida...