| --- |
| base_model: |
| - Qwen/Qwen2.5-VL-7B-Instruct |
| language: |
| - en |
| license: apache-2.0 |
| pipeline_tag: image-text-to-text |
| tags: |
| - transformers |
| - multimodal |
| library_name: transformers |
| --- |
| |
| # VL-Rethinker-7B |
|
|
| **🚀 News:** <u>We release our meticulously curated collection of RL training queries for multimodal reasoning: [ViRL39K](https://huggingface.co/datasets/TIGER-Lab/ViRL39K).</u> |
|
|
| **VL-Rethinker-7B** achieves SoTA results on various multimodal reasoning benchmarks. |
|
|
| It is trained using the **GRPO-SSR and Forced Rethinking** techniques, using meticulously curated [ViRL39K](https://huggingface.co/datasets/TIGER-Lab/ViRL39K). |
|
|
| For details of our approach and performance comparison, please see our [paper](https://github.com/TIGER-AI-Lab/VL-Rethinker/blob/main/paper.pdf). |
|
|
| For details of training and evaluation, please see our [code repo](https://github.com/TIGER-AI-Lab/VL-Rethinker/). |
|
|
| Explore further via the following links: |
|
|
| | [**🚀Project Page**](https://tiger-ai-lab.github.io/VL-Rethinker/) | [**📖Paper**](https://arxiv.org/abs/2504.08837) | [**🔗Github**](https://github.com/TIGER-AI-Lab/VL-Rethinker/) | [**🤗Data**](https://huggingface.co/datasets/TIGER-Lab/ViRL39K) | |
|
|
| ## Citation |
|
|
| If you feel this model useful, please give us a free cite: |
| ```bibtex |
| @article{vl-rethinker, |
| title={VL-Rethinker: Incentivizing Self-Reflection of Vision-Language Models with Reinforcement Learning}, |
| author = {Wang, Haozhe and Qu, Chao and Huang, Zuming and Chu, Wei and Lin,Fangzhen and Chen, Wenhu}, |
| journal={arXiv preprint arXiv:2504.08837}, |
| year={2025} |
| } |
| ``` |
