diffusion-reasoning/gdsd_sudoku_dream

GDSD: Guided Denoiser Self-Distillation for Diffusion Language Models

This repository contains a model checkpoint from the paper GDSD: Reinforcement Learning as Guided Denoiser Self-Distillation for Diffusion Language Models.

Guided Denoiser Self-Distillation (GDSD) is a reinforcement learning framework that improves the denoiser of diffusion large language models (dLLMs) by distilling from an advantage-guided self-teacher. This approach bypasses the biases of traditional ELBO-based methods and provides more stable training dynamics for dLLMs across planning, math, and coding benchmarks.

Links

• —Paper: https://arxiv.org/abs/2605.29398
• —Code: https://github.com/GaryBall/GDSD

Citation

If you find GDSD helpful, please consider citing:

@misc{tang2026gdsdreinforcementlearningguided,
      title={GDSD: Reinforcement Learning as Guided Denoiser Self-Distillation for Diffusion Language Models}, 
      author={Xiaohang Tang and Keyue Jiang and Che Liu and Qifang Zhao and Xiaoxiao Xu and Sangwoong Yoon and Ilija Bogunovic},
      year={2026},
      eprint={2605.29398},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2605.29398}, 
}