---
license: apache-2.0
library_name: transformers
base_model: Qwen/Qwen2.5-VL-3B-Instruct
base_model_relation: finetune
pipeline_tag: image-text-to-text
---

<a href="https://arxiv.org/abs/2610.12403" target="_blank">
  <img alt="arXiv" src="https://img.shields.io/badge/arXiv-ViSkill-red?logo=arxiv" height="20" />
</a>
<a href="https://github.com/ZJU-REAL/ViSkill" target="_blank">
  <img alt="Code" src="https://img.shields.io/badge/Code-ViSkill-white?logo=github" height="20" />
</a>
<a href="https://huggingface.co/hongxingli/ViSkill-Sokoban" target="_blank">
  <img alt="ViSkill-Sokoban" src="https://img.shields.io/badge/Model-ViSkill--Sokoban-ffc107?logo=huggingface" height="20" />
</a>

# ViSkill-FrozenLake

This repository contains **ViSkill-FrozenLake** and its visual skill library, introduced in [ViSkill: Reinforcing VLM Agents with Evolving Visual-Native Skills](https://arxiv.org/abs/2610.12403).

## Model Description

ViSkill-FrozenLake is a visual agent model for FrozenLake, built upon [Qwen2.5-VL-3B-Instruct](https://huggingface.co/Qwen/Qwen2.5-VL-3B-Instruct). It represents successful interactions as reusable visual skill cards. Geometry-aware retrieval guides the agent, while PPO with skill-guided rewards and online skill distillation jointly improve the policy and skill library. Model files are stored at the repository root, with the accompanying visual skill library in `skill_library/`.

## Usage

```python
from transformers import AutoProcessor, Qwen2_5_VLForConditionalGeneration

repo_id = "hongxingli/ViSkill-FrozenLake"

model = Qwen2_5_VLForConditionalGeneration.from_pretrained(repo_id)
processor = AutoProcessor.from_pretrained(repo_id)
```

For the full ViSkill agent pipeline, including skill retrieval and environment interaction, please refer to our [code repository](https://github.com/ZJU-REAL/ViSkill).

## Citation

```bibtex
@misc{li2026viskillreinforcingvlmagents,
      title={ViSkill: Reinforcing VLM Agents with Evolving Visual-Native Skills},
      author={Hongxing Li and Dingming Li and Yixin Li and Yong Du and Wenqi Zhang and Weiming Lu and Jun Xiao and Yueting Zhuang and Yongliang Shen},
      year={2026},
      eprint={2610.12403},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2610.12403},
}
```