一作ACL 2026Findings· Findings of ACL
Think Smart, Not Hard: Difficulty Adaptive Reasoning for Large Audio Language Models
Zhichao Sheng · Shilin Zhou · Chen GONG · Zhenghua Li
摘要
Large Audio Language Models (LALMs) employing the Chain-of-Thought paradigm have demonstrated remarkable reasoning capabilities. Though different problems naturally require varying depths of reasoning, existing methods often determine whether to perform reasoning, lacking fine-grained mechanisms...