Thyme-RL
收藏魔搭社区2026-01-06 更新2025-08-23 收录
下载链接:
https://modelscope.cn/datasets/Kwai-Keye/Thyme-RL
下载链接
链接失效反馈官方服务:
资源简介:
<div align="center">
<img src="https://cdn-uploads.huggingface.co/production/uploads/685ba798484e3233f5ff6f11/dxBp6TmwqwNuBuJR9gfQC.png" width="40%" alt="Thyme Logo">
</div>
<font size=4><div align='center' >
[[📖 Home Page](https://thyme-vl.github.io/)]
[[📖 Github Repo](https://github.com/yfzhang114/Thyme)]
[[📖 Technique Report](https://arxiv.org/abs/2508.11630)]
[[📊 Thyme SFT Model](https://huggingface.co/Kwai-Keye/Thyme-SFT)]
[[📊 Thyme RL Model](https://huggingface.co/Kwai-Keye/Thyme-RL)]
[[📝 SFT Data](https://huggingface.co/datasets/Kwai-Keye/Thyme-SFT)]
[[📝 RL Data](https://huggingface.co/datasets/Kwai-Keye/Thyme-RL)]
</div></font>
## 🔥 News
* **`2025.08.15`** 🌟 We are excited to introduce **Thyme: Think Beyond Images**. Thyme transcends traditional ``thinking with images'' paradigms by autonomously generating and executing diverse image processing and computational operations through executable code, significantly enhancing performance on high-resolution perception and complex reasoning tasks. Leveraging a novel two-stage training strategy that combines supervised fine-tuning with reinforcement learning and empowered by the innovative GRPO-ATS algorithm, Thyme achieves a sophisticated balance between reasoning exploration and code execution precision.
<div align="center">
<img src="https://cdn-uploads.huggingface.co/production/uploads/685ba798484e3233f5ff6f11/c_D7uX3RT1WUANDRB70ZC.png" width="100%" alt="Thyme Logo">
</div>
We have provided the usage instructions, training code, and evaluation code in the [GitHub repo](https://github.com/yfzhang114/Thyme).
## Citation
If you find Thyme useful in your research or applications, please cite our paper:
```bibtex
@misc{zhang2025thymethinkimages,
title={Thyme: Think Beyond Images},
author={Yi-Fan Zhang and Xingyu Lu and Shukang Yin and Chaoyou Fu and Wei Chen and Xiao Hu and Bin Wen and Kaiyu Jiang and Changyi Liu and Tianke Zhang and Haonan Fan and Kaibing Chen and Jiankang Chen and Haojie Ding and Kaiyu Tang and Zhang Zhang and Liang Wang and Fan Yang and Tingting Gao and Guorui Zhou},
year={2025},
eprint={2508.11630},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2508.11630},
}
```
<div align="center">
<img src="https://cdn-uploads.huggingface.co/production/uploads/685ba798484e3233f5ff6f11/dxBp6TmwqwNuBuJR9gfQC.png" width="40%" alt="Thyme 标志">
</div>
<font size=4><div align='center' >
[[📖 主页](https://thyme-vl.github.io/)]
[[📖 GitHub 仓库](https://github.com/yfzhang114/Thyme)]
[[📖 技术报告](https://arxiv.org/abs/2508.11630)]
[[📊 Thyme-SFT 模型](https://huggingface.co/Kwai-Keye/Thyme-SFT)]
[[📊 Thyme-RL 模型](https://huggingface.co/Kwai-Keye/Thyme-RL)]
[[📝 SFT 数据集](https://huggingface.co/datasets/Kwai-Keye/Thyme-SFT)]
[[📝 RL 数据集](https://huggingface.co/datasets/Kwai-Keye/Thyme-RL)]
</div></font>
## 🔥 最新动态
* **`2025.08.15`** 🌟 我们荣幸地推出**Thyme:超越图像的思考**。Thyme突破了传统“以图像为基础进行思考”的范式,可通过可执行代码自主生成并执行多样化的图像处理与计算操作,显著提升了高分辨率感知与复杂推理任务的性能。该模型采用创新的两阶段训练策略,结合监督微调(Supervised Fine-Tuning, SFT)与强化学习(Reinforcement Learning, RL),并依托开创性的GRPO-ATS算法,在推理探索与代码执行精度之间达成了精妙的平衡。
<div align="center">
<img src="https://cdn-uploads.huggingface.co/production/uploads/685ba798484e3233f5ff6f11/c_D7uX3RT1WUANDRB70ZC.png" width="100%" alt="Thyme 标志">
</div>
我们已在[GitHub 仓库](https://github.com/yfzhang114/Thyme)中提供了使用说明、训练代码与评估代码。
## 引用格式
如果您的研究或应用场景中用到了Thyme,请引用我们的论文:
bibtex
@misc{zhang2025thymethinkimages,
title={Thyme: Think Beyond Images},
author={Yi-Fan Zhang and Xingyu Lu and Shukang Yin and Chaoyou Fu and Wei Chen and Xiao Hu and Bin Wen and Kaiyu Jiang and Changyi Liu and Tianke Zhang and Haonan Fan and Kaibing Chen and Jiankang Chen and Haojie Ding and Kaiyu Tang and Zhang Zhang and Liang Wang and Fan Yang and Tingting Gao and Guorui Zhou},
year={2025},
eprint={2508.11630},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2508.11630},
}
提供机构:
maas
创建时间:
2025-08-17
搜集汇总
数据集介绍

背景与挑战
背景概述
Thyme-RL数据集是Thyme项目的一部分,专注于通过可执行代码进行图像处理和计算操作,采用了两阶段训练策略和GRPO-ATS算法,以提高高分辨率感知和复杂推理任务的性能。数据集包含使用说明、训练和评估代码,适用于研究和应用场景。
以上内容由遇见数据集搜集并总结生成



