Qwen3-4B-Ins-Draft-OPD

This repository contains Qwen3-4B-Ins-Draft-OPD, a draft model for speculative decoding.

The model is post-trained from z-lab/Qwen3-4B-DFlash-b16. It keeps the overall architecture and inference interface consistent with the original DFlash draft model, while further adapting the draft model through the Draft-OPD post-training method.

• Paper: Draft-OPD: On-Policy Distillation for Speculative Draft Models
• Project Page: https://www.haodilei.top/draft-opd/
• Code: https://github.com/bingyang-lei/Draft-OPD

Model Details

• Base draft model: Qwen3-4B(enable_thinking=true)
• Model type: Draft model for speculative decoding
• Architecture: Same as the original DFlash draft model
• Post-training method: Draft-OPD (On-Policy Distillation)

Method Summary

Draft-OPD trains speculative draft models with on-policy target feedback. Instead of only learning from fixed target-generated trajectories (SFT), the drafter is supervised on draft-induced states exposed during speculative verification. This allows the drafter to learn from target feedback on both accepted and rejected proposals, focusing training on the draft-induced errors that limit speculative acceptance.

For detailed training procedures, evaluation settings, and performance results, please refer to the paper.

Citation

If you find our work useful, please consider citing our paper:

@misc{lei2026draftopdonpolicydistillationspeculative,
      title={Draft-OPD: On-Policy Distillation for Speculative Draft Models}, 
      author={Haodi Lei and Yafy Li and Haoran Zhang and Shunkai Zhang and Qianjia Cheng and Xiaoye Qu and Ganqu Cui and Bowen Zhou and Ning Ding and Yun Luo and Yu Cheng},
      year={2026},
      eprint={2605.29343},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2605.29343}, 
}