---
base_model:
- Qwen/Qwen2.5-VL-7B-Instruct
datasets:
- WaltonFuture/Multimodal-Cold-Start
- WaltonFuture/Multimodal-RL-Data
license: apache-2.0
pipeline_tag: image-text-to-text
library_name: transformers
---

* 🐙 **GitHub Repo:** [waltonfuture/RL-with-Cold-Start](https://github.com/waltonfuture/RL-with-Cold-Start)
* 📜 **Paper (arXiv):** [Advancing Multimodal Reasoning via Reinforcement Learning with Cold Start (arXiv:2505.22334)](https://arxiv.org/abs/2505.22334)

<div align=center>
<img src="assets/model_comparison.png"  width = "80%" alt="model_comparison" align=center/>
</div>

## Cold Start Stage

We conduct supervised fine-tuning on Qwen2.5-VL-3B and Qwen2.5-VL-7B using [ms-swift](https://github.com/modelscope/ms-swift). In this stage, please refer to this curated [dataset](https://huggingface.co/datasets/WaltonFuture/Multimodal-Cold-Start) distilled from Qwen2.5-VL-32B using rejection sampling.

### Setup

```bash
git clone https://github.com/waltonfuture/RL-with-Cold-Start.git
cd RL-with-Cold-Start/SFT
pip install -e .
```

### Prepare Data

```bash
python convert_data.py
```

### SFT

```bash
bash qwen2.5vl_sft.sh
```

The checkpoint can be found in SFT/output.

## RL Stage

We further conduct GRPO using [EasyR1](https://github.com/hiyouga/EasyR1). Please refer to this [dataset](https://huggingface.co/datasets/WaltonFuture/Multimodal-RL-Data) for the GRPO training.

### Setup

```bash
git clone https://github.com/waltonfuture/RL-with-Cold-Start.git
cd RL-with-Cold-Start/GRPO
pip install -e .
```

### GRPO Training (replace the checkpoint with the model after SFT)

```bash
bash examples/qwen2_5_vl_7b_grpo.sh
```

### Merge Checkpoint in Hugging Face Format

```bash
python3 scripts/model_merger.py --local_dir checkpoints/easyr1/qwen2_5_vl_7b_grpo/global_step_80/actor
```

## Data Access

Our two stage datasets are now available on Huggingface.

|  Stage          | Data  |
| ------------------ | ------------- |
| Cold Start  | [Multimodal-Cold-Start](https://huggingface.co/datasets/WaltonFuture/Multimodal-Cold-Start) |
| RL  | [Multimodal-RL-Data](https://huggingface.co/datasets/WaltonFuture/Multimodal-RL-Data) |

## Model Access

Our models are now available on Huggingface.

|  Backbone           | Our model   |
| ------------------ | ------------- |
| Qwen2.5-VL-7b  | [Qwen2.5VL-7b-RL-with-Cold-Start](https://huggingface.co/WaltonFuture/Qwen2.5VL-7b-RLCS) |
| Qwen2.5-VL-3b  | [Qwen2.5VL-3b-RL-with-Cold-Start](https://huggingface.co/WaltonFuture/Qwen2.5VL-3b-RLCS) |

## Acknowledgment

Our models are built upon the amazing [Qwen2.5-VL](https://huggingface.co/collections/Qwen/qwen25-vl-6795ffac22b334a837c0f9a5) family.
We thank [EasyR1](https://github.com/hiyouga/EasyR1) and [ms-swift](https://github.com/modelscope/ms-swift) for their training codes.

## Contact

Please contact Lai Wei (waltonfuture@sjtu.edu.cn) if needed.

## Citation
```
@article{wei2025advancing,
  title={Advancing Multimodal Reasoning via Reinforcement Learning with Cold Start},
  author={Wei, Lai and Li, Yuting and Zheng, Kaipeng and Wang, Chen and Wang, Yue and Kong, Linghe and Sun, Lichao and Huang, Weiran},
  journal={arXiv preprint arXiv:2505.22334},
  year={2025}
}
```