MOPD: Multi-Teacher On-Policy Distillation for Capability Integration in LLM Post-Training
For LLM practitioners, MOPD provides an efficient and effective method to combine multiple specialized skills into one model without cross-domain coupling, enabling parallel development of domain teachers.
MOPD introduces a multi-teacher on-policy distillation paradigm that integrates multiple domain-specific capabilities into a single LLM, outperforming baselines like Mix-RL and Cascade RL on Qwen3-30B-A3B by inheriting nearly all teacher capabilities.
Modern large language models (LLMs) rely on reinforcement learning during post-training to push specific capabilities, yet integrating multiple capabilities into one model remains hard. Existing methods, such as Off-Policy Finetune and Mix-RL, are either inefficient or lose performance. In this work, we propose Multi-teacher On-Policy Distillation (MOPD), a post-training paradigm for combining the capabilities of multiple domain RL teachers: we first run per-domain specialised RL to obtain a set of domain teachers, then distill these teachers into the student on its own rollouts. This eliminates exposure bias and provides a dense optimization signal. On Qwen3-30B-A3B, MOPD outperforms Mix-RL, Cascade RL, Off-Policy Finetune, and Param-Merge baselines, inheriting nearly all of each teacher's capability. MOPD also enables parallel, independent development of domain teachers, removing the cross-domain coupling typical of multi-domain post-training. MOPD has been deployed in the post-training of MiMo-V2-Flash, an industrial-scale frontier model, demonstrating its practical value for capability integration in frontier-scale LLMs.