---
license: mit
base_model:
- meta-llama/Meta-Llama-3-8B-Instruct
datasets:
- HuggingFaceH4/ultrachat_200k
- walledai/HarmBench
language:
- en
new_version: ASSELab/DAT-Llama-3-8B-Instruct
tags:
- pytorch
- llama
- llama-3
- DAT
- robust
- adversarial
library_name: transformers
paper:
  title: "Closing the Distribution Gap in Adversarial Training for LLMs"
  url: "https://arxiv.org/abs/2602.15238"
---

# DAT - Distributional Adversarial Training

[![arXiv](https://img.shields.io/badge/arXiv-2602.15238-b31b1b.svg)](https://arxiv.org/abs/2602.15238)
[![GitHub](https://img.shields.io/badge/GitHub-DAT-181717?logo=github&logoColor=white)](https://github.com/ASSELab/DAT)

DAT utilizes [continuous adversarial training](https://arxiv.org/abs/2405.15589) on [diffusion-based](https://arxiv.org/abs/2511.00203v1) adversarial examples to close the gap between empirical and population-robust risk.
We fine-tune [meta-llama/Meta-Llama-3-8B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3-8B-Instruct).

For further information, consult our paper [https://arxiv.org/abs/2602.15238](https://arxiv.org/abs/2602.15238) or repository [https://github.com/ASSELab/DAT](https://github.com/ASSELab/DAT)

## Citation

```tex
@misc{hu2026closingdistributiongapadversarial,
      title={Closing the Distribution Gap in Adversarial Training for LLMs}, 
      author={Chengzhi Hu and Jonas Dornbusch and David Lüdke and Stephan Günnemann and Leo Schwinn},
      year={2026},
      eprint={2602.15238},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2602.15238}, 
}
```