The paper introduces UECR-GRPO, a method that unifies on‑policy distillation and verifier‑based reinforcement learning for mathematical reasoning. It combines verifier rewards and teacher‑derived log‑ratios into a single KL‑regularized objective (Path‑Utility Unification) and then redistributes credit at the token level using entropy‑calibrated redistribution, preserving total task credit. Experiments on five benchmarks show that UECR‑GRPO improves average accuracy by up to 0.89 percentage points over the best baseline for both Qwen3‑1.7B and Qwen3‑4B students.
By Jie Zhang, Jingxiao Yang, Zhehao Huang, Yuhang Liu, Xiaolin Huang
The paper proposes a reward‑density principle for language‑model post‑training: sparse sequence‑level rewards are best for exploring and discovering new behaviors in a strong teacher, while dense token‑level teacher supervision is optimal for compressing that behavior into a smaller deployment model. The authors validate this rule with a four‑stage workflow—teacher RL, forward‑KL warmup, on‑policy distillation, and optional post‑bridge student RL—showing that an RL‑improved 8B teacher distilled through a dense bridge outperforms direct GRPO on a 1.7B student across math and code benchmarks. Ablation studies confirm each stage’s contribution, and the approach generalizes to Llama and Qwen models, offering a practical guideline for allocating scarce labeled data.
By Yuanda Xu, Hejian Sang, Zhengze Zhou, Ran He, Zhipeng Wang, Alborz Geramifard
The paper introduces a recursive self-improvement framework for language models that replaces an external teacher with a frozen copy of the student, enabling dynamic co-evolution (DCE) and self-refined concise learning (SRCL). DCE allows the privileged teacher to evolve alongside the student, while SRCL trains on shorter, verified rewrites to reduce verbosity. Experiments show that the combined DCE+SRCL approach outperforms traditional on‑policy self‑distillation across multiple model sizes and math benchmarks, achieving significant accuracy gains and shorter outputs.
By Shangjian Yin, Zehao Zhao, Kavosh Asadi, Rui Liu, Yuchen Lu, Shike Mei, Hang Cui, Luke Simon, Zhouxing Shi, Hamed Firooz
arXiv:2607. 05394v1 Announce Type: cross Abstract: Reinforcement learning with verifiable rewards (RLVR) is a powerful recipe for improving language-model reasoning, but it is expensive to repeat on every new strong model because the target model must generate many rollouts during training.
By Shiyuan Feng, Huan-ang Gao, Haohan Chi, Hanlin Wu, Zhilong Zhang, Zheng Jiang, Bingxiang He, Wei-Ying Ma, Ya-Qin Zhang, Hao Zhou
arXiv:2607. 29209v1 Announce Type: cross Abstract: Reinforcement learning with verifiable rewards (RLVR) broadcasts a single response-level reward to every token, while on-policy distillation (OPD) scores each token against a stronger teacher for a dense advantage but caps performance at teacher quality and discourages exploration beyond it.
By Yifan Ding, Xincheng Wei, Yoshua Y. Li, Ziheng Li, Yuquan Lu, Siyu Zhang, Dongsheng Ma, Rongxiang Weng, Xunliang Cai, Yun Chen
Reinforcement learning with verifiable rewards (RLVR) is a powerful recipe for improving language-model reasoning, but it is expensive to repeat on every new strong model because the target model must generate many rollouts during training. As models scale, post-training itself becomes a bottleneck.
arXiv:2607. 18955v1 Announce Type: new Abstract: Reinforcement learning with verifiable rewards (RLVR) has substantially improved the reasoning capabilities of large language models on tasks such as mathematical reasoning and code generation.
By Qiye Cai, Yichuan Ma, Linyang Li, Peiji Li, Yongkang Chen, Qipeng Guo, Yicheng Zou, Tao Gui, Xiaocheng Feng, Bing Qin
The paper introduces VISTA, a method that enhances on‑policy self‑distillation (OPSD) by adapting the teacher model toward the student’s distribution using outcome‑verified rollouts. VISTA keeps the standard OPSD student update but selectively adjusts the teacher only on the top‑k positions with the largest teacher‑student KL divergence, without adding new sampling or reward objectives. Experiments on AIME24, AIME25, and HMMT25 with Qwen3 models show that VISTA outperforms OPSD across all scales, improving Avg@12 by up to 2.1 points.
By Zewen Ding, Zezhong Wu, Zhou Tao, Shida Wang, Shizhuo Hou, YongXiang Hua, Haoyu Cao, Linli Xu
The paper investigates on‑policy distillation (OPD), showing that teacher supervision during OPD contains significant noise that grows with teacher size, yet the student policy remains largely unaffected by this noise. It finds that OPD’s gains stem mainly from suppressing low‑log‑probability tokens, a process that can be replicated without a teacher. Building on this insight, the authors propose On‑Policy Self‑Adaptation (OPSA), a supervision‑free method that uses entropy‑adaptive negative advantages to improve performance on several benchmarks, outperforming both the base model and OPD.
By Yi Ding, Ruqi Zhang
arXiv:2607. 04037v1 Announce Type: cross Abstract: On-policy distillation is a powerful way to transfer reasoning ability from a strong teacher to a smaller student: the student samples trajectories from its own policy, and the teacher provides dense token-level supervision on the states the student actually visits.
By Mohammad Sadegh Akhondzadeh, Vijay Lingam, Atula Tejaswi, Chanakya Ekbote, Sujay Sanghavi, Aleksandar Bojchevski
arXiv:2609.27421v1 Announce Type: new
Abstract: Multi-constraint instruction following requires a model to respond to a query under many simultaneously active constraints. Even strong instruction-tun...
By Yanzhao Zheng, Yuanqiang Yu, Tianze Xu, Chao Ma, Zhentao Zhang, Jihuai Zhu, Baohua Dong, Hangcheng Zhu, Ruohui Huang
The paper introduces Joint On-Policy Learning and Teaching (JOLT), a method that trains a single policy to act both as a privileged teacher and an unprivileged student. It derives a condition ensuring the teacher’s distillation update aligns positively with the student’s reward gradient, leading to a KL‑regularized teacher objective that matches the student’s current capabilities. Experiments show JOLT improves training efficiency and performance across various tasks, with additional gains when student rewards are incorporated.
By Randy Ardywibowo, Arnav Dalal, Jiantao Jiao