作者位次 #4/8ACL 2026OralOutstanding Paper· Natural Language Processing
Evolutionary Guided Decoding: Iterative Value Refinement for LLM s
Zhenhua Liu · Lijun Li · Ruizhe Chen · Yuxian Jiang · Tong Zhu · Zhaochen Su · Wenliang Chen · Jing Shao
摘要
While guided decoding, especially value-guided methods, has emerged as a cost-effective alternative for controlling language model outputs without re-training models, its effectiveness is limited by the accuracy of the value function. We identify that this inaccuracy stems from a core distributio...