Co-Evolving Policy Distillation

TL;DR

This paper introduces Co-Evolving Policy Distillation (CoPD), a method that co-trains multiple experts bidirectionally to better consolidate diverse capabilities in a single model, outperforming existing paradigms.

Contribution

The paper proposes CoPD, a novel co-evolutionary training approach that integrates expert capabilities more effectively than traditional sequential or unidirectional methods.

Findings

CoPD outperforms mixed RLVR and MOPD baselines in integrating capabilities.

CoPD surpasses domain-specific experts in experiments.

CoPD enables a new training paradigm for scalable, multi-capability models.

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.