---
license: apache-2.0
library_name: transformers
pipeline_tag: image-text-to-text
---

# Model Card for AtomThinkPRM

The model is fine-tuned with atomic step execution based on math-psa and can be used for process supervision in multimodal reasoning chains. It is part of the **AtomThink** framework, which introduces "slow thinking" into multimodal large language models (MLLMs).

- **Paper:** [AtomThink: Multimodal Slow Thinking with Atomic Step Reasoning](https://huggingface.co/papers/2411.11930)
- **Repository:** [https://github.com/Kun-Xiang/AtomThink](https://github.com/Kun-Xiang/AtomThink)

## Description

AtomThink incorporates the notion of "slow thinking" into MLLMs, allowing models to adaptively use different levels of reasoning for questions of varying complexity. It proposes a novel paradigm of Self-structured Chain of Thought (SCoT), which consists of minimal semantic atomic steps. 

AtomThinkPRM is designed for process supervision, enabling the evaluation of single-step reasoning quality within these multimodal chains.

# Citation
If you use this model in your research, please cite:

```bibtex
@article{xiang2024atomthink,
  title={AtomThink: A Slow Thinking Framework for Multimodal Mathematical Reasoning},
  author={Xiang, Kun and Liu, Zhili and Jiang, Zihao and Nie, Yunshuang and Huang, Runhui and Fan, Haoxiang and Li, Hanhui and Huang, Weiran and Zeng, Yihan and Han, Jianhua and others},
  journal={arXiv preprint arXiv:2411.11930},
  year={2024}
}

@article{wang2024openr,
  title={OpenR: An Open Source Framework for Advanced Reasoning with Large Language Models},
  author={Wang, Jun and Fang, Meng and Wan, Ziyu and Wen, Muning and Zhu, Jiachen and Liu, Anjie and Gong, Ziqin and Song, Yan and Chen, Lei and Ni, Lionel M and others},
  journal={arXiv preprint arXiv:2410.09671},
  year={2024}
}
```

# License
The checkpoint is released under the Apache 2.0 license. Please ensure proper attribution when using this checkpoint.