The paper investigates how on-policy self‑distillation can alter a model’s behavior by conditioning on privileged information. It contrasts attractive self‑distillation, which pulls a model toward a privileged teacher, with repulsive self‑distillation, which pushes it away, showing that attraction reduces exploratory reasoning while repulsion lengthens responses and can destabilize the model. The authors propose a contrastive self‑distillation objective that combines attraction to a correct‑solution teacher with repulsion from an incorrect‑solution teacher, finding that this approach improves reasoning performance across various model types while keeping response lengths stable.
arXiv:2606. 11709v1 Announce Type: new Abstract: On-policy self-distillation (OPSD) provides dense, token-level supervision for reasoning models by aligning a model's own distribution with the distribution it produces under privileged context, typically a verified solution.
arXiv:2605. 11458v3 Announce Type: replace Abstract: On-policy self-distillation has become a strong recipe for LLM reasoning, where a privileged teacher supervises the student's own rollouts while conditioning on the reference solution.
arXiv:2608. 04794v1 Announce Type: new Abstract: Self-distillation (SD) has emerged as a compute-efficient alternative to reinforcement learning with verifiable rewards: a self-teacher, conditioned on privileged information (PI) about the answer such as a reference solution, supplies dense per-token supervision to a student that never sees it.
On-policy self-distillation (OPSD) provides dense, token-level supervision for reasoning models by aligning a model's own distribution with the distribution it produces under privileged context, typically a verified solution. However, we show that the learning signal drawn from this distributional gap concentrates on style tokens rather than task-bearing ones, as the hinted model tends to produce more direct, shorter outputs.
arXiv:2606. 11173v1 Announce Type: new Abstract: Conditioning a language model on additional context, such as feedback on a previous attempt, typically improves its response.
arXiv:2607. 05184v1 Announce Type: new Abstract: Self-distillation is a promising recipe for self-improvement in language models.
On-policy distillation trains a language model on its own generations while a teacher scores them token by token. It combines the dense supervision of imitation learning with the on-policy sampling of...
The paper reviews On‑Policy Self‑Distillation (OPSD), a method where a language model learns from its own generations using privileged information such as reference solutions or plans, eliminating the need for a larger teacher model. It identifies a key failure mode—collapse, where the model’s reasoning paths narrow progressively—and analyzes it through three levers: signal application, privileged information, and teacher dynamics. The review focuses on mathematical reasoning, offering a unified vocabulary and distinguishing settled facts from ongoing debates.
The paper investigates self‑distillation techniques for language models by systematically varying three key design choices: the source of rollout tokens (student vs. teacher), the teacher coupling strategy (frozen or exponential moving average), and the KL divergence direction (reverse or forward). Experiments on Qwen2.5‑7B and Ministral‑3‑3B across 1,200 adaptation runs reveal that rollout source mainly affects acquisition on contradictory tasks, teacher coupling most strongly influences acquisition across all tasks, and KL direction impacts retention differently depending on the model. A controlled theoretical model reproduces these empirical trends, offering a unified framework for understanding acquisition‑retention trade‑offs in self‑distillation.
arXiv:2608. 09826v1 Announce Type: cross Abstract: Reinforcement learning with verifiable rewards yields no group-relative signal when rollout groups are uniformly correct or uniformly wrong, which account for 63.
Reinforcement learning with verifiable rewards yields no group-relative signal when rollout groups are uniformly correct or uniformly wrong, which account for 63. 0-68.
The paper investigates how privileged information—such as a teacher’s full solution or reasoning trace—affects on‑policy self‑distillation (OPSD) in language models. Using the AMPLE‑Math benchmark, the authors compare distillation with and without extra teacher views, finding that reference‑free distillation explains most gains for Qwen3‑1.7B, while additional references provide modest benefits, especially for polished solutions. The study also shows that the impact of privileged data depends on the student’s training regime and that altering token‑level supervision can leave student behavior largely unchanged.