CORE: Context-Robust Remasking for Diffusion Language Models

Kevin Zhai, Sabbir Mollah, Zhenyi Wang, Mubarak Shah
University of Central Florida

bash run_all.sh

Citation

If you use this code or find the paper useful, please cite:

@article{zhai2026corecontextrobustremaskingdiffusion,
      title={CORE: Context-Robust Remasking for Diffusion Language Models}, 
      author={Kevin Zhai and Sabbir Mollah and Zhenyi Wang and Mubarak Shah},
      year={2026},
      eprint={2602.04096},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2602.04096}, 
}

Acknowledgements

We thank colleagues and collaborators for discussions and feedback that improved this work. We also acknowledge the authors and maintainers of the open-source libraries, pretrained models (e.g., LLaDA), and evaluation suites (e.g., lm-eval) used in this repository, along with the creators of the benchmarks and datasets used in our experiments. Finally, we appreciate the compute infrastructure and operational support that enabled the runs reported here.