作者位次 #2/4ACL 2026Findings· Findings of ACL
Hidden States as Early Signals: Step-level Trace Evaluation and Pruning for Efficient Test-Time Scaling
Zhixiang Liang · Beichen Huang · Zheng Wang · Minjia Zhang
摘要
Large Language Models (LLMs) can enhance reasoning capabilities through test-time scaling by generating multiple traces. However, the combination of lengthy reasoning traces with multiple sampling introduces substantial computation and high end-to-end latency. Prior work on accelerating this proc...