diffusion-reasoning/gdsd_countdown_llada

GDSD: Reinforcement Learning as Guided Denoiser Self-Distillation for Diffusion Language Models

This repository contains the model weights for GDSD, as presented in the paper GDSD: Reinforcement Learning as Guided Denoiser Self-Distillation for Diffusion Language Models.

Guided Denoiser Self-Distillation (GDSD) is a reinforcement learning (RL) framework tailored for diffusion large language models (dLLMs). It improves performance by directly distilling the denoiser from an advantage-guided self-teacher, bypassing the biases typically found in evidence lower bound (ELBO) based likelihood surrogates.

Links

• —Paper: arXiv:2605.29398
• —GitHub Repository: https://github.com/GaryBall/GDSD

Citation

If you find this work 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}, 
}