作者位次 #3/7ACL 2026Findings· Findings of ACL
D iffu S pec: Unlocking Diffusion Language Models for Speculative Decoding
Guanghao Li · Zhihui Fu · Min Fang · Qibin Zhao · Ming Tang · Chun Yuan · Jun Wang
摘要
Autoregressive (AR) decoding in large language models (LLMs) is latency-bounded by strictly sequential token generation.Speculative decoding mitigates this bottleneck by letting a fast drafter propose multi-token candidates that are then verified in parallel by the target model; yet most existing...