| title | Co-Evolving Policy Distillation |
|---|---|
| authors | Naibin Gu, Chenxu Yang, Qingyi Si, Chuanyu Qin, Dingyu Yao, Peng Fu, Zheng Lin, Weiping Wang, Nan Duan, Jiaqi Wang |
| date | 2026-04-29 |
| venue | arXiv (preprint, not peer reviewed) |
| id | arXiv:2604.27083 |
| upvotes | 67 (Hugging Face Daily Papers, 2026-11-01) |
| links | abs pdf hf |
| abstract | RLVR and OPD have become standard paradigms for post-training. We provide a unified analysis of these two paradigms in consolidating multiple expert capabilities into a single model, identifying capability loss in different ways: mixed RLVR suffers from inter-capability divergence cost, while the pipeline of first training experts and then performing OPD, though avoiding divergence, fails to fully absorb teacher capabilities due to large behavioral pattern gaps between teacher and student. We propose Co-Evolving Policy Distillation (CoPD), which encourages parallel training of experts and introduces OPD during each expert's ongoing RLVR training rather than after complete expert training, with experts serving as mutual teachers (making OPD bidirectional) to co-evolve. This enables more consistent behavioral patterns among experts while maintaining sufficient complementary knowledge throughout. Experiments validate that CoPD achieves all-in-one integration of text, image, and video reasoning capabilities, significantly outperforming strong baselines such as mixed RLVR and MOPD, and even surpassing domain-specific experts. The model parallel training pattern offered by CoPD may inspire a novel training scaling paradigm. |
cite
@misc{gu2026co,
title={Co-Evolving Policy Distillation},
author={Naibin Gu and Chenxu Yang and Qingyi Si and Chuanyu Qin and Dingyu Yao and Peng Fu and Zheng Lin and Weiping Wang and Nan Duan and Jiaqi Wang},
year={2026},
eprint={2604.27083},
archivePrefix={arXiv},
url={https://arxiv.org/abs/2604.27083}
}