paper-with-me

Papers

OPSD Compresses What RLVR Teaches: A Post-RL Compaction Stage for Reasoning Models

2026-05-07 · Jaehoon Kim, Dongha Lee arxiv

On-Policy Self-Distillation (OPSD) has recently emerged as an alternative to Reinforcement Learning with Verifiable Rewards (RLVR), promising higher accuracy and shorter responses through token-level credit assignment from a self-teacher conditioned on privileged context. However, this promise does not carry over to thinking-enabled mathematical reasoning, where reported accuracy gains shrink and sometimes turn negative. We hypothesize that hindsight supervision can specify better token-level alternatives in short thinking-disabled outputs, but in long thinking-enabled traces it more readily identifies redundancy than supplies better replacements. To test this, we applied OPSD separately to correct and incorrect rollout groups, so that compression and correction can be observed in isolation. Our results show that in thinking-enabled mathematical reasoning, OPSD behaves most reliably as a compression mechanism rather than a correction mechanism: training only on correct rollouts preserves accuracy while substantially shortening responses, whereas training only on incorrect rollouts damages accuracy. In light of these findings, we propose a revised post-training pipeline for thinking-enabled mathematical reasoning: SFT then RLVR then OPSD.

📄 PDF Abstract BibTeX arXiv:2605.06188

Code (0)

등록된 구현이 없습니다.

Tasks

Reinforcement LearningMathematical Reasoning

Similar Papers 제목 키워드 기반

Learning from the Self-future: On-policy Self-distillation for dLLMs

2026-06-16 · Yifu Luo, Zeyu Chen, Haoyu Wang, Xinhao Hu 외 arxiv

On-policy self-distillation (OPSD) has proven effective for post-training large language models (LLMs), yet its application to diffusion LLMs (dLLMs) remains unexplored. Existing OPSD methods are inherently autoregressiv…

ReDraft, Don't Just Distill: Reference-Driven Revision for Continual VLLM Post-Training

2026-09-15 · Zhihao Zhang, Mingqi Wu, Qiaole Dong, Enyu Zhou 외 arxiv

Continual post-training of large multimodal models should add new capabilities while preserving those from pre-training, and the two goals pull in opposite directions. SFT gives explicit target supervision that learns a …

The Many Faces of On-Policy Distillation: Pitfalls, Mechanisms, and Fixes

2026-05-11 · Siqi Zhu, Xuyan Ye, Hongyu Lu, Weiye Shi 외 arxiv

On-policy distillation (OPD) and on-policy self-distillation (OPSD) have emerged as promising post-training methods for large language models, offering dense token-level supervision on trajectories sampled from the model…

Mathematical Reasoning

Contrastive Reinforced Policy Optimization via Privileged Self-Distillation

2026-07-30 · Xingjian Wu, Junlin Liu, Xingchen Liu, Xuhang Zhu 외 arxiv

Recent advances in post-training Large Language Models (LLMs) increasingly rely on Reinforcement Learning with Verifiable Rewards (RLVR) or On-Policy Self-Distillation (OPSD). While OPSD provides dense, logit-level super…

Reinforcement LearningContrastive Learning

Self-Distilled RLVR

2026-04-03 · Chenxu Yang, Chuanyu Qin, Qingyi Si, Minghui Chen 외 arxiv

On-policy distillation (OPD) has become a popular training paradigm in the LLM community. This paradigm selects a larger model as the teacher to provide dense, fine-grained signals for each sampled trajectory, in contras…

Reinforcement Learning