Saved in:
| Main Authors: | , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2510.02816 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866908574465130496 |
|---|---|
| author | Zhang, Yulong Wang, Li Du, Wei Li, Peilin Zhao, Yuqin Dai Zhiyuan Fang, Lingyong Liu, Ziniu Zhang, Ru Zhu, Huijia Liu, Gongshen |
| author_facet | Zhang, Yulong Wang, Li Du, Wei Li, Peilin Zhao, Yuqin Dai Zhiyuan Fang, Lingyong Liu, Ziniu Zhang, Ru Zhu, Huijia Liu, Gongshen |
| contents | Verifying multi-step reasoning in large language models is difficult due to imprecise error localization and high token costs. Existing methods either assess entire reasoning chains, suffering attention dilution, or rely on expensive multi-sampling. We introduce Node-wise Consistency Verification (NCV), a training-free framework that recasts verification as lightweight binary consistency checks at the node level. By decomposing the chain of thought into interconnected verification nodes, NCV precisely localizes errors and avoids unnecessary long-form generation. Experiments demonstrate that our approach enhances interpretability and efficiency, presenting a scalable solution for reliable LLM reasoning verification. On public datasets, NCV achieves a 10\% to 25\% improvement in F1 scores over baselines while utilizing $6\times$~$58\times$ fewer tokens than traditional methods like CoT-based verifiers. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2510_02816 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | NCV: A Node-Wise Consistency Verification Approach for Low-Cost Structured Error Localization in LLM Reasoning Zhang, Yulong Wang, Li Du, Wei Li, Peilin Zhao, Yuqin Dai Zhiyuan Fang, Lingyong Liu, Ziniu Zhang, Ru Zhu, Huijia Liu, Gongshen Artificial Intelligence Computation and Language Verifying multi-step reasoning in large language models is difficult due to imprecise error localization and high token costs. Existing methods either assess entire reasoning chains, suffering attention dilution, or rely on expensive multi-sampling. We introduce Node-wise Consistency Verification (NCV), a training-free framework that recasts verification as lightweight binary consistency checks at the node level. By decomposing the chain of thought into interconnected verification nodes, NCV precisely localizes errors and avoids unnecessary long-form generation. Experiments demonstrate that our approach enhances interpretability and efficiency, presenting a scalable solution for reliable LLM reasoning verification. On public datasets, NCV achieves a 10\% to 25\% improvement in F1 scores over baselines while utilizing $6\times$~$58\times$ fewer tokens than traditional methods like CoT-based verifiers. |
| title | NCV: A Node-Wise Consistency Verification Approach for Low-Cost Structured Error Localization in LLM Reasoning |
| topic | Artificial Intelligence Computation and Language |
| url | https://arxiv.org/abs/2510.02816 |