metadata
library_name: transformers
license: apache-2.0
pipeline_tag: video-text-to-text
This repository contains the GRPO-CARE model, presented in the paper GRPO-CARE: Consistency-Aware Reinforcement Learning for Multimodal Reasoning. GRPO-CARE is a novel consistency-aware RL framework that jointly optimizes for both answer correctness and reasoning coherence, without requiring explicit process supervision. It introduces a two-tiered reward system to address limitations of standard outcome-supervised GRPO, improving both accuracy and logical consistency in multimodal reasoning.
Code released at GRPO-CARE.
Citation
@misc{chen2025grpocareconsistencyawarereinforcementlearning,
title={GRPO-CARE: Consistency-Aware Reinforcement Learning for Multimodal Reasoning},
author={Yi Chen and Yuying Ge and Rui Wang and Yixiao Ge and Junhao Cheng and Ying Shan and Xihui Liu},
year={2025},
eprint={2506.16141},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2506.16141},
}