Stabilizing On-Policy Distillation for MLLM Reasoning with Global Normalization
About
On-policy distillation (OPD) has recently emerged as an important post-training paradigm. By using a stronger teacher model to provide dense, fine-grained supervision for sampled trajectories, OPD offers a clear advantage over reinforcement learning with verifiable rewards (RLVR), which typically depends on sparse binary or outcome-based environmental feedback. However, naive token-level distillation can suffer from gradient instability, due to magnitude misalignment in outlier states. To address this issue, we propose Globally Normalized Distillation Policy Optimization (GNDPO), a practical method that stabilizes optimization by transforming raw KL scores into batch-level relative advantages. This normalization effectively mitigates gradient explosions while retaining the benefits of token-level guidance. Experimental results show that GNDPO substantially improves training robustness and downstream performance across multimodal reasoning tasks. The code is released at https://github.com/OPPO-Mente-Lab/GNDPO.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Multimodal Math Reasoning | MathVision | Accuracy51.6 | 263 | |
| Multimodal Math Reasoning | WeMath | Accuracy48.8 | 228 | |
| Multimodal Mathematical Reasoning | MathVista mini | Accuracy0.73 | 124 | |
| Multimodal Logical Reasoning | LogicVista | Accuracy54.4 | 76 | |
| Massive Multi-discipline Multimodal Understanding | MMMU (val) | -- | 18 | |
| Multimodal Mathematical Reasoning | MathVerse Vision Only | Accuracy53.2 | 13 | |
| Multimodal Mathematical Reasoning | DynaMath worst case | Accuracy32.5 | 13 |