The paper introduces Zone of Proximal Policy Optimization (ZPPO), a method that keeps a teacher model inside prompts rather than in the policy gradient to improve knowledge distillation for small students. ZPPO creates two types of reformulated prompts—Binary Candidate-included Questions (BCQ) and Negative Candidate-included Questions (NCQ)—to expose students to correct and incorrect responses, and uses a replay buffer to focus training on hard questions until the student’s accuracy improves. Experiments on the Qwen3.5 family with a 27B teacher across 31 benchmarks show that ZPPO outperforms both off‑policy and on‑policy distillation methods, especially at the smallest student scales.
On-policy learning is central to post-training: training on a model's own generations limits distribution shift and helps preserve existing capabilities. On hard problems, however, on-policy RL provid...
arXiv:2610.27467v1 Announce Type: cross Abstract: On-policy learning is central to post-training: training on a model's own generations limits distribution shift and helps preserve existing capabilit...
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.
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:2608. 09836v1 Announce Type: new Abstract: On-policy distillation (OPD) has emerged as a core component of modern LLM post-training pipelines, yet we reveal a failure mode: degenerate agreement, where students exploit repetitive loops to achieve near-perfect token agreement with the teacher despite globally flawed responses.
arXiv:2609.17474v1 Announce Type: cross Abstract: Large language model (LLM) distillation aims to transfer the capabilities of a powerful teacher to a smaller student. Direct imitation, however, can...
arXiv:2607. 26246v1 Announce Type: new Abstract: On-policy distillation (OPD), which aligns a student with the teacher's token-level distribution on the student's own rollouts, is an effective paradigm for transferring capabilities across LLMs.
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 Residual Advantage (RA), a new guidance method for reinforcement learning with verifiable rewards (RLVR) that treats the teacher–student probability residual as a bounded one‑step reward, subtracts the student’s state value to form an advantage, and centers it within each response. RA preserves the verifier’s mean label while redistributing credit among response steps, and its extension CoRA updates the teacher model using verifier advantages to adapt guidance to the student’s attempts. Experiments with Qwen3 models show that RA combined with GRPO or REINFORCE++ consistently outperforms teacher‑only on‑policy distillation across three mathematical benchmarks, improving macro Avg@8 and Pass@8 scores, with CoRA providing additional gains.
The paper examines on‑policy self‑distillation, where a teacher scores a student’s rollout to provide dense token‑level credit. Experiments show that the raw scoring signals neither localize errors nor effectively penalize incorrect tokens, because locating errors is a search problem that the teacher is not explicitly asked to perform. Introducing a critique‑conditioned approach—first generating a critique of the rollout and then conditioning the teacher on it—produces signals that accurately localize errors, correct erroneous spans, and guide the policy toward backtracking tokens, improving performance on math problems and in cold‑start GRPO settings.
The paper introduces Multi-Teacher Self-Distillation Policy Optimization (MT‑SDPO), an on‑policy distillation method that combines multiple frozen teachers into a single student model. MT‑SDPO uses self‑anchors, answer‑verified eligibility, and privileged distillation to select reliable teachers per sample rather than per domain. Experiments on five students from three model families show that MT‑SDPO improves the weakest domain of Qwen3‑8B by 14.79 points and reduces its domain gap by 74.7%, achieving a more balanced performance than matching a single teacher to each domain.
arXiv:2607. 27770v1 Announce Type: new Abstract: A single reinforcement-learning run can produce a strong reasoner yet an incomplete teacher: it often amplifies only a subset of the valid solution modes.