paper-with-me

Papers

Multi-Rollout On-Policy Distillation via Peer Successes and Failures

2026-05-12 · Weichen Yu, Xiaomin Li, Yizhou Zhao, Xiaoze Liu, Ruowang Zhang, Haixin Wang, Yinyi Luo, Chen Henry Wu, Gaurav Mittal, Matt Fredrikson, Yu Hu arxiv

Large language models are often post-trained with sparse verifier rewards, which indicate whether a sampled trajectory succeeds but provide limited guidance about where reasoning succeeds or fails. On-policy distillation (OPD) offers denser token-level supervision by training on student-generated trajectories, yet existing methods typically distill each rollout independently and ignore the other attempts sampled for the same prompt. We introduce Multi-Rollout On-Policy Distillation (MOPD), a peer-conditioned distillation framework that uses the student's local rollout group to construct more informative teacher signals. MOPD conditions the teacher on both successful and failed peer rollouts: successes provide positive evidence for valid reasoning patterns, while failures provide structured negative evidence about plausible mistakes to avoid. We study two peer-context constructions: positive peer imitation and contrastive success-failure conditioning. Experiments on competitive programming, mathematical reasoning, scientific question answering, and tool-use benchmarks show that MOPD consistently improves over standard on-policy baselines. Further teacher-signal analysis shows that mixed success-failure contexts better align teacher scores with verifier rewards, indicating that the gains arise from more faithful, instance-adaptive supervision. These results indicate that effective on-policy distillation should exploit the student's multi-rollout trial-and-error behavior rather than treating rollouts as isolated samples.

📄 PDF Abstract BibTeX arXiv:2605.12652

Code (0)

등록된 구현이 없습니다.

Tasks

Mathematical ReasoningQuestion Answering

Similar Papers 제목 키워드 기반

Localizing Credit at the Divergence: Path-Conditioned Self-Distillation for LLM Reasoning

2026-06-14 · Yu Li, Shu Hong, Tian Lan arxiv

Reinforcement learning from verifiable rewards assigns a single scalar to each rollout, leaving token-level credit assignment underspecified in long reasoning traces. On-policy self-distillation addresses this by letting…

Reinforcement Learning

Be My Tutor: On-Policy Co-Distillation for Mutual LLM Improvement via Peer Feedback

2026-06-12 · Woohyeon Byeon, Jiwon Jeon, Jeonghye Kim, Youngchul Sung arxiv

We study multi-domain LLM training in which two models, each stronger in a different domain, co-evolve by tutoring each other through on-policy feedback. Unlike one-way distillation or single-model fine-tuning, our goal …

Experience Sharing in Mutual Reinforcement Learning for Heterogeneous Language Models

2026-05-08 · Xiaoze Liu, Dhananjay Ram, Yuting Zhang, Zhaoyang Zhang 외 arxiv

We introduce Mutual Reinforcement Learning, a framework for concurrent RL post-training in which heterogeneous LLM policies exchange typed experience while keeping separate parameters, objectives, and tokenizers. The fra…

Reinforcement Learning

Trust-Region Behavior Blending for On-Policy Distillation

2026-05-29 · Daniil Plyusov, Alexey Gorbatovski, Alexey Malakhov, Nikita Balagansky 외 arxiv

On-policy distillation (OPD) trains a student on prefixes sampled from its own policy while matching a stronger teacher. This addresses the prefix mismatch of offline distillation, but early student rollouts can still be…

Rethinking On-Policy Distillation of Large Language Models II: One Training Example

2026-09-03 · Zixuan Fu, Bingxiang He, Yuxin Zuo, Haohuan Huang 외 hf

On-policy distillation (OPD) combines student-generated rollouts with dense token-level supervision from a teacher. Existing work has mainly studied its algorithmic behavior, leaving the role of training data unclear. We…