Omanic: Towards Step-wise Evaluation of Multi-hop Reasoning in Large Language Models 事件

Omanic: Towards Step-wise Evaluation of Multi-hop Reasoning in Large Language Models arXiv:2603.16654v2 Announce Type: replace Abstract: Evaluating the reasoning abilities of large language models (LLMs) solely from final answers can obscure failures in intermediate steps, especially in multi-hop QA benchmarks without step-level annotations. To address this gap, we introduce Omanic, an open-domain 4-hop QA benchmark designed not only to measure final-answer accuracy but also to diagnose where r