作者位次 #5/8ACL 2026Findings· Findings of ACL
TAR o: Token-level Adaptive Routing for LLM Test-time Alignment
Arushi Rai · Qiang Zhang · Hanqing Zeng · Yunkai Zhang · Dipesh Tamboli · Xiangjun Fan · Zhuokai Zhao · Lizhu Zhang
摘要
Large language models (LLMs) exhibit strong reasoning capabilities but typically require expensive post-training to reach high performance. Recent test-time alignment methods offer a lightweight alternative, but have been explored mainly for preference alignment rather than reasoning. To bridge t...