Stabilizing On-Policy Distillation for MLLM Reasoning with Global Normalization

• URL: http://arxiv.org/abs/2606.09091v1
• Date: Mon, 08 Jun 2026 06:41:31 GMT
• Title: Stabilizing On-Policy Distillation for MLLM Reasoning with Global Normalization
• Abstract summary: On-policy distillation (OPD) has recently emerged as an important post-training paradigm.<n>We propose Globally Normalized Distillation Policy Optimization (GNDPO), a practical method that stabilizes optimization by transforming raw KL scores into batch-level relative advantages.
• Score: 9.860371691501076
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: 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 papers

• Distill Where You Fail: Recovering Learning Signals of Negative RL-Groups from Adaptive Teacher Guidance [50.77757355236912]
  We propose RSTG (Recovering Learning Signals via Adaptive Teacher Guidance), which applies distillation selectively and precisely where it matters most.<n> Experiments demonstrate that RSTG substantially outperforms naive GRPO+OPD by +4.02% on math and+3.05% on code.
  arXiv  Detail & Related papers  (2026-08-01T17:29:14Z)
• One-Way Policy Optimization for Self-Evolving LLMs [63.8638342097375]
  Reinforcement Learning with Verifiable Rewards (RLVR) has become a promising paradigm for scaling reasoning capabilities of Large Language Models (LLMs)<n>We propose One-Way Policy Optimization (OWPO), a method based on the principle of decoupling optimization direction from update magnitude.<n> Experimental results demonstrate that OWPO outperforms strong baselines, including DAPO, OPD, and MOPD.
  arXiv  Detail & Related papers  (2026-05-21T08:25:27Z)
• KL for a KL: On-Policy Distillation with Control Variate Baseline [13.281263788199219]
  On-Policy Distillation (OPD) has emerged as a dominant post-training paradigm for large language models.<n>We propose vOPD, which casts OPD as policy-gradient RL.<n>We show that vOPD consistently outperforms vanilla OPD and matches the most expensive full-vocabulary baseline.
  arXiv  Detail & Related papers  (2026-05-08T15:24:51Z)
• Asymmetric On-Policy Distillation: Bridging Exploitation and Imitation at the Token Level [23.59808224440603]
  On-policy distillation (OPD) trains a student on its own trajectories with token-level teacher feedback and often outperforms off-policy distillation and standard reinforcement learning.<n>We find that its standard advantage weighted policy gradient suffers from three structural weaknesses, including high variance updates, vanishing gradients in zero-advantage regions, and exploration bottlenecks when corrective signals are insufficient.<n>We propose Asymmetric On-Policy Distillation (AOPD), which replaces ineffective negative reinforcement with localized divergence minimization in non-positive advantage regions while preserving positive reinforcement learning.
  arXiv  Detail & Related papers  (2026-05-07T15:02:49Z)
• EP-GRPO: Entropy-Progress Aligned Group Relative Policy Optimization with Implicit Process Guidance [3.463914032107119]
  We propose Entropy-Progress Aligned GRPO, a framework that mines the model's intrinsic information flow for dense, self-supervised guidance.<n>EP- GRPO achieves superior accuracy and efficiency compared to GRPO and its variants.
  arXiv  Detail & Related papers  (2026-05-06T14:21:54Z)
• EVPO: Explained Variance Policy Optimization for Adaptive Critic Utilization in LLM Post-Training [69.32453275232662]
  We show that a learned critic can inject estimation noise that exceeds the state signal it captures, increasing rather than reducing advantage variance.<n>We propose Explained Variance Policy Optimization (EVPO), which monitors batch-level EV at each training step and adaptively switches between critic-based and batch-mean advantage estimation.
  arXiv  Detail & Related papers  (2026-04-21T14:07:39Z)
• Rethinking the Trust Region in LLM Reinforcement Learning [72.25890308541334]
  Proximal Policy Optimization (PPO) serves as the de facto standard algorithm for Large Language Models (LLMs)<n>We propose Divergence Proximal Policy Optimization (DPPO), which substitutes clipping with a more principled constraint.<n>DPPO achieves superior training and efficiency compared to existing methods, offering a more robust foundation for RL-based fine-tuning.
  arXiv  Detail & Related papers  (2026-02-04T18:59:04Z)
• A Step Back: Prefix Importance Ratio Stabilizes Policy Optimization [58.116300485427764]
  Reinforcement learning post-training can elicit reasoning behaviors in large language models.<n> token-level correction often leads to unstable training dynamics when the degree of off-policyness is large.<n>We propose a simple yet effective objective, Minimum Prefix Ratio (MinPRO)
  arXiv  Detail & Related papers  (2026-01-30T08:47:19Z)
• Margin Adaptive DPO: Leveraging Reward Model for Granular Control in Preference Optimization [0.0]
  Margin-Adaptive Direct Preference Optimization provides a stable, data-preserving, and instance-level solution.<n>We provide a comprehensive theoretical analysis, proving that MADPO has a well-behaved optimization landscape.<n>It achieves performance gains of up to +33.3% on High Quality data and +10.5% on Low Quality data over the next-best method.
  arXiv  Detail & Related papers  (2025-10-06T20:09:37Z)
• From Noisy Traces to Stable Gradients: Bias-Variance Optimized Preference Optimization for Aligning Large Reasoning Models [90.45197506653341]
  Large reasoning models generate intermediate reasoning traces before producing final answers.<n> aligning LRMs with human preferences, a crucial prerequisite for model deployment, remains underexplored.<n>A common workaround optimized a single sampled trajectory, which introduces substantial gradient variance from trace sampling.
  arXiv  Detail & Related papers  (2025-10-06T17:58:01Z)
• Stabilizing Policy Gradients for Sample-Efficient Reinforcement Learning in LLM Reasoning [77.92320830700797]
  Reinforcement Learning has played a central role in enabling reasoning capabilities of Large Language Models.<n>We propose a tractable computational framework that tracks and leverages curvature information during policy updates.<n>The algorithm, Curvature-Aware Policy Optimization (CAPO), identifies samples that contribute to unstable updates and masks them out.
  arXiv  Detail & Related papers  (2025-10-01T12:29:32Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.