作者位次 #2/8ACL 2026Poster· Natural Language Processing
Calibrating Inference Time Alignment with Sequence-level Risk Accumulation
Shanwen Tan · Ziyang Dong · Wei Ju · Yiwei Fu · Hao Wu · Kun Wang · Yifan Wang · Ziyue Qiao
摘要
This paper investigates the problem of safe decoding for Large Language Models (LLMs) during inference, particularly under jailbreak attacks. Previous approaches typically either detect malicious content or regulate the decoding alignment of LLMs to mitigate such attacks. Although effective in de...