titleP^2-MOPD: Accelerating Multi-Teacher On-Policy Distillation via Parameter Priors
authorsSinuo Wang, Jiacheng Lu, Zeyu Zhang, Maojia Song, Rui Sun, Beidi Luan, Yutong Xie, Jing Li, Daxin Jiang, Qi Wu, Zuo Bai
date2026-11-02
venuearXiv (preprint, not peer reviewed)
idarXiv:2611.03953
affiliationStepFun
upvotes51 (Hugging Face Daily Papers, 2026-11-11)
linksabs pdf hf code
cite
relatedStepAudio 3 Realtime Technical Report
StepAudio 3 Music Technical Report
onPanda: Efficient Annotation of On-Policy Alignment Data for LLMs and Agents via Token-Level Correction
ShutterMuse: Capture-Time Photography Guidance with MLLMs
Φ-Bench: Can Large Language Models Engineer the Infrastructure That Powers Them?
abstractMulti-teacher on-policy distillation consolidates expert capabilities by supervising student-generated trajectories, whereas task arithmetic offers a training-free alternative through linear parameter combinations. We view the linear combination of student and expert parameters as a single differentiable model and propose P^2-MOPD, which jointly optimizes the mixing coefficients and the student weights under the same on-policy distillation objective. Unlike task arithmetic, which stays within the span of the frozen student and expert parameters for any choice of merging coefficients, P^2-MOPD keeps the student trainable, so the experts act as parameter priors without confining it to this span. Across four domains at 1.7B and 4B scales, P^2-MOPD achieves faster and more balanced recovery than MOPD from the same initialization, surpassing its final average recovery in half the training steps. At the end of training, our 1.7B model reaches 94.7% average recovery, versus 80.9% for MOPD.

bibtex

@misc{wang2026p,
  title={P^2-MOPD: Accelerating Multi-Teacher On-Policy Distillation via Parameter Priors},
  author={Sinuo Wang and Jiacheng Lu and Zeyu Zhang and Maojia Song and Rui Sun and Beidi Luan and Yutong Xie and Jing Li and Daxin Jiang and Qi Wu and Zuo Bai},
  year={2026},
  eprint={2611.03953},
  archivePrefix={arXiv},
  url={https://arxiv.org/abs/2611.03953}
}