Enregistré dans:
| Auteurs principaux: | , , , , , , |
|---|---|
| Format: | Preprint |
| Publié: |
2026
|
| Sujets: | |
| Accès en ligne: | https://arxiv.org/abs/2603.08251 |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866911498969808896 |
|---|---|
| author | Zhang, Dongxu Lin, Hongqiang Sun, Yiding Wang, Pengyu Wang, Qirui Yang, Ning Zhu, Jihua |
| author_facet | Zhang, Dongxu Lin, Hongqiang Sun, Yiding Wang, Pengyu Wang, Qirui Yang, Ning Zhu, Jihua |
| contents | Scaling test-time computation enhances LLM reasoning ability but faces a uniform computation paradox. Allocating identical resources leads to over-correction on simple tasks and insufficient refinement on complex ones. To address this, we propose CoFiCot, a coarse-to-fine adaptive framework that dynamically tailors inference strategies to problem difficulty. Specifically, we implement a multi-metric classifier that triages queries by synthesizing semantic entropy, consensus reliability, and predicted reasoning depth . This enables a differentiated refinement stage that applies efficient aggregation for simple queries while routing complex ones to a context-aware correction loop . We formalize correction as a stateful sequential propagation process , where each repair is strictly conditioned on the verified history of prior rectifications. By integrating Process Reward Models (PRMs) within this state-dependent trajectory, CoFiCot effectively bridges the gap between granular error localization and global logical coherence, preventing the context fragmentation typical of stateless refinement methods. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2603_08251 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Not All Queries Need Deep Thought: CoFiCot for Adaptive Coarse-to-fine Stateful Refinement Zhang, Dongxu Lin, Hongqiang Sun, Yiding Wang, Pengyu Wang, Qirui Yang, Ning Zhu, Jihua Computation and Language Scaling test-time computation enhances LLM reasoning ability but faces a uniform computation paradox. Allocating identical resources leads to over-correction on simple tasks and insufficient refinement on complex ones. To address this, we propose CoFiCot, a coarse-to-fine adaptive framework that dynamically tailors inference strategies to problem difficulty. Specifically, we implement a multi-metric classifier that triages queries by synthesizing semantic entropy, consensus reliability, and predicted reasoning depth . This enables a differentiated refinement stage that applies efficient aggregation for simple queries while routing complex ones to a context-aware correction loop . We formalize correction as a stateful sequential propagation process , where each repair is strictly conditioned on the verified history of prior rectifications. By integrating Process Reward Models (PRMs) within this state-dependent trajectory, CoFiCot effectively bridges the gap between granular error localization and global logical coherence, preventing the context fragmentation typical of stateless refinement methods. |
| title | Not All Queries Need Deep Thought: CoFiCot for Adaptive Coarse-to-fine Stateful Refinement |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2603.08251 |