---
license: apache-2.0
base_model:
- Qwen/Qwen3-4B-Instruct
---

## Model Description

**Mem-T-4B** refers to the model parameters derived from training Qwen3-4B-Instruct using **MoT-GRPO** within the **Mem-T** framework.



## Usage

For detailed instructions on how to use within the **Mem-T** framework, please refer to the main [Mem-T GitHub repository](https://github.com/yanweiyue/Mem-T).


## Links

* **GitHub:** [https://github.com/yanweiyue/Mem-T](https://github.com/yanweiyue/Mem-T)

## Citation

If you find this work useful, please consider citing our paper.

```
@misc{yue2026memtdensifyingrewardslonghorizon,
      title={Mem-T: Densifying Rewards for Long-Horizon Memory Agents}, 
      author={Yanwei Yue and Guibin Zhang and Boci Peng and Xuanbo Fan and Jiaxin Guo and Qiankun Li and Yan Zhang},
      year={2026},
      eprint={2601.23014},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2601.23014}, 
}
```