Bridging Perception and Reasoning: Token Reweighting for RLVR in Multimodal LLMs
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , , , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2026
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866914423058202624 |
|---|---|
| author | Lu, Jinda Wu, Junkang Li, Jinghan Huang, Kexin Yang, Shuo Wang, Guoyin Wu, Jiancan Wang, Xiang He, Xiangnan |
| author_facet | Lu, Jinda Wu, Junkang Li, Jinghan Huang, Kexin Yang, Shuo Wang, Guoyin Wu, Jiancan Wang, Xiang He, Xiangnan |
| contents | Extending Reinforcement Learning with Verifiable Rewards (RLVR) to multimodal large language models (MLLMs) faces a fundamental challenge: their responses inherently interleave perception-related tokens, which ground visual content, with reasoning-related tokens, which construct reasoning chains. These token types instantiate distinct yet interdependent capacities -- visual grounding and symbolic reasoning -- making isolated optimization insufficient. Through token-level empirical analysis, we demonstrate that optimizing either perception- or reasoning-only tokens consistently underperforms full optimization, underscoring their inherent coupling. To address this, we propose a plug-and-play Token-Reweighting (ToR) strategy that explicitly models this interdependence by identifying critical tokens of both types and dynamically reweighting them during RLVR training. Applied on top of existing methods (e.g., GRPO and DAPO), ToR delivers consistent performance gains across multiple multi-modal reasoning benchmarks, achieving state-of-the-art performance with both accurate visual grounding and coherent reasoning. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2603_25077 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Bridging Perception and Reasoning: Token Reweighting for RLVR in Multimodal LLMs Lu, Jinda Wu, Junkang Li, Jinghan Huang, Kexin Yang, Shuo Wang, Guoyin Wu, Jiancan Wang, Xiang He, Xiangnan Computer Vision and Pattern Recognition Extending Reinforcement Learning with Verifiable Rewards (RLVR) to multimodal large language models (MLLMs) faces a fundamental challenge: their responses inherently interleave perception-related tokens, which ground visual content, with reasoning-related tokens, which construct reasoning chains. These token types instantiate distinct yet interdependent capacities -- visual grounding and symbolic reasoning -- making isolated optimization insufficient. Through token-level empirical analysis, we demonstrate that optimizing either perception- or reasoning-only tokens consistently underperforms full optimization, underscoring their inherent coupling. To address this, we propose a plug-and-play Token-Reweighting (ToR) strategy that explicitly models this interdependence by identifying critical tokens of both types and dynamically reweighting them during RLVR training. Applied on top of existing methods (e.g., GRPO and DAPO), ToR delivers consistent performance gains across multiple multi-modal reasoning benchmarks, achieving state-of-the-art performance with both accurate visual grounding and coherent reasoning. |
| title | Bridging Perception and Reasoning: Token Reweighting for RLVR in Multimodal LLMs |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2603.25077 |