DualOPSD introduces an asymmetric alternating framework for on-policy self-distillation, where a privileged teacher first trains the student and then adapts to the student’s updated distribution on the same trajectory, eliminating the need for additional rollouts. This adaptive approach improves average performance on AIME 2024, AIME 2025, and HMMT 2025 benchmarks for the Qwen3-8B model, with gains that vary by model scale. Across 1.7B, 4B, and 8B scales, DualOPSD also reduces truncation effects and lowers KL divergence between teacher and student.
arXiv:2608.26019v2 Announce Type: replace-cross Abstract: On-policy self-distillation (OPSD) combines student generated trajectories with dense supervision from a privileged teacher of the same model...
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.
arXiv:2608. 01735v2 Announce Type: replace Abstract: On-policy (self) distillation (OPSD) is increasingly adopted for language-model post-training.
The paper investigates on‑policy distillation (OPD), a method that trains a student model using its own rollouts while a teacher remains frozen, to see if it can help a student improve on problems it currently fails. Experiments with Qwen‑3 and Phi‑4 models on a competition‑mathematics benchmark show that OPD actually degrades performance when training focuses on failure cases, even though training loss (KL divergence) continues to drop. However, when OPD is applied to a broader set of problems, it improves accuracy by 4.6 percentage points, indicating that OPD works well when most training data contains correct student prefixes but can reinforce errors if the training regime concentrates on failures.
RetireOPD introduces a self-retiring on‑policy distillation method for agentic reinforcement learning. It first trains a skill‑conditioned teacher with environment rewards, then jointly trains a skill‑free student with RL and OPD, allowing the student to autonomously stop using the teacher when its performance aligns with the teacher’s. Experiments on Qwen2.5 models show significant gains in ALFWorld success rates and WebShop accuracy compared to RL baselines and the teacher itself.
arXiv:2610.19145v1 Announce Type: new Abstract: On-policy distillation (OPD) is increasingly studied as a cheaper alternative to reinforcement learning, yet it remains a form of knowledge distillatio...
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.
The paper introduces Teacher-Gated On-Policy Distillation (TGOPD), a method that verifies teacher reliability at the prompt level before applying dense supervision in on-policy distillation. TGOPD uses verifier-scored teacher probes to decide whether to route a prompt to dense OPD or to a verifier-grounded alternative. Experiments on 4B and 35B models across mathematics, code, and instruction tasks show TGOPD outperforms vanilla OPD and improves teacher GPU utilization from 9.8% to 78.9% in a 4B single-domain run.
arXiv:2609.36546v1 Announce Type: cross Abstract: On-policy distillation (OPD) trains a student model on its self-generated trajectories with dense token-level teacher feedback. However, naive OPD ma...
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.
arXiv:2607. 18293v1 Announce Type: new Abstract: On-policy self-distillation (OPSD) teaches large language models new skills through a teacher that shares the student's backbone and supervises its own rollouts.
arXiv:2609.08798v1 Announce Type: new Abstract: Weak-to-strong generalization asks whether stronger models can learn from weaker supervisors and surpass them. This question is particularly important...