This repository is for our paper:

dVoting: Fast Voting for dLLMs
Sicheng Feng¹, Zigeng Chen¹, Xinyin Ma¹, Gongfan Fang¹, Xinchao Wang^1,*
1National University of Singapore, Singapore
^∗Corresponding author: xinchao@nus.edu.sg

---

🙋 Please let us know if you find a mistake or have any suggestions!

🌟 If you find this resource helpful, please consider to star this repository and cite our research!

Updates

• 2026-02-13: 🚀 arXiv online!
• 2026-02-12: 🚀 We release code for dVoting!

Usage

1. Install dependencies

If you face any issues with the installation, please feel free to open an issue. We will try our best to help you.

pip install -r requirements.txt

2. Download datasets and models

You can download GSM8K & MATH500 & GPQA & ARC-C & MMLU / LLaDA-8B-Instruct & GSAI-ML/LLaDA-1.5 & Dream-org/Dream-v0-Instruct-7B for evaluation by directly running the scripts or from HuggingFace.

3. Evaluation

We follow simple-evals to evaluate our method on the benchmarks. We provide some scripts to facilitate the evaluation.

# baseline (majority voting or HEX)
bash scripts/run_baseline.sh # llada / llada1.5
bash scripts/run_baseline_dream.sh # dream

# dVoting
bash scripts/run_dvoting.sh # llada / llada1.5
bash scripts/run_dvoting_dream.sh # dream

Acknowledgement

We thank the authors of LLaDA, Dream, dKV-Cache, simple-evals for open-sourcing their codebases.

Citation

If you find this paper useful in your research, please consider citing our papers:

@article{feng2026dvoting,
      title={dVoting: Fast Voting for dLLMs}, 
      author={Feng, Sicheng and Chen, Zigeng and Ma, Xinyin and Fang, Gongfan and Wang, Xinchao},
      journal={arXiv preprint arXiv:2602.12153},
      year={2026},
}