# When Clipping Reverses Correction: Failure Dynamics of Pointwise Forward-KL On-Policy Self-Distillation

arXiv: https://arxiv.org/abs/2609.38995 · HTML: https://mlsanity.com/paper/2609.38995/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Placed in On-policy and self-distillation for LLMs, one of 12 directions in Knowledge distillation.
- Builds on 23 of its 27 references that are in the map.
- First posted on arXiv 30 Sep 2026.

## Placed in

- Knowledge distillation / [On-policy and self-distillation for LLMs](https://mlsanity.com/distill/on-policy-and-self-distillation-for-llms/)

## Builds on (most cited first, 23 of 23)

- [Qwen3 Technical Report](https://mlsanity.com/paper/2505.09388/) (2025)
- [HybridFlow: A Flexible and Efficient RLHF Framework](https://mlsanity.com/paper/2409.19256/) (2024)
- [On-Policy Distillation of Language Models: Learning from Self-Generated Mistakes](https://mlsanity.com/paper/2306.13649/) (2023)
- [Self-Distilled Reasoner: On-Policy Self-Distillation for Large Language Models](https://mlsanity.com/paper/2601.18734/) (2026)
- [Reinforcement Learning via Self-Distillation](https://mlsanity.com/paper/2601.20802/) (2026)
- [OpenThoughts: Data Recipes for Reasoning Models](https://mlsanity.com/paper/2506.04178/) (2025)
- [Self-Distillation Enables Continual Learning](https://mlsanity.com/paper/2601.19897/) (2026)
- [RLCSD: Reinforcement Learning with Contrastive On-Policy Self-Distillation](https://mlsanity.com/paper/2606.11709/) (2026)
- [When Are Teacher Tokens Reliable? Position-Weighted On-Policy Self-Distillation for Reasoning](https://mlsanity.com/paper/2605.21606/) (2026)
- [On-Policy Self-Distillation without Any Supervision](https://mlsanity.com/paper/2608.06296/) (2026)
- [OGLS-SD: On-Policy Self-Distillation with Outcome-Guided Logit Steering for LLM Reasoning](https://mlsanity.com/paper/2605.12400/) (2026)
- [Rethinking Reward Supervision: Rubric-Conditioned Self-Distillation](https://mlsanity.com/paper/2606.19327/) (2026)
- [Privileged Solutions or Context-Induced Teacher Behavior? Dissecting On-Policy Self-Distillation](https://mlsanity.com/paper/2608.09228/) (2026)
- [Self-Supervised On-Policy Distillation for Reasoning Language Models](https://mlsanity.com/paper/2605.17497/) (2026)
- [Search-E1: Self-Distillation Drives Self-Evolution in Search-Augmented Reasoning](https://mlsanity.com/paper/2605.22511/) (2026)
- [Is More Privileged Information Better? From Solution Traces to Problem-Solving Structure in Self-Distilled Reasoning](https://mlsanity.com/paper/2608.01589/) (2026)
- [Rethinking Privileged Information in On-Policy Self-Distillation](https://mlsanity.com/paper/2608.18271/) (2026)
- [PAINT: Partial-Solution Adaptive Interpolated Training for Self-Distilled Reasoners](https://mlsanity.com/paper/2604.26573/) (2026)
- [DASH: Divergence-Adaptive Supervision Horizons for On-Policy Self-Distillation of Reasoning Models](https://mlsanity.com/paper/2608.06243/) (2026)
- [PAST: Privileged Adaptation from Complete Student Trajectories for On-Policy Self-Distillation](https://mlsanity.com/paper/2608.08726/) (2026)
- [DualOPSD: Adaptive Privileged Teachers for On-Policy Self-Distillation](https://mlsanity.com/paper/2608.26019/) (2026)
- [What Does Privileged Information Add to On-Policy Self-Distillation?](https://mlsanity.com/paper/2609.20612/) (2026)
- [SCOPE-OPSD: Fisher-Conditioned Privileged Subspaces for On-Policy Self-Distillation](https://mlsanity.com/paper/2609.12579/) (2026)

## Built on by (most cited first, 1 of 1)

- [Learning from Self-Generated Responses: Reliability--Uncertainty Self-Distillation for Reasoning](https://mlsanity.com/paper/2610.19176/) (2026)

## Abstract (arXiv)

On-policy self-distillation (OPSD) trains a student on its own generated responses using feedback from the same model conditioned on privileged information. On mathematical reasoning, the original OPSD study finds that stylistic tokens can dominate the training signal over math-related tokens, and that pointwise clipping of the forward KL objective stabilizes training. Pointwise clipping caps each vocabulary-wise forward KL term at a fixed threshold before summing over the vocabulary. Follow-up studies have adopted this clipping, but its effect on training has not been directly examined. In matched training runs differing only in whether clipping is applied, we observe that clipped runs produce substantially more repetitions that persist to the end of the response than their unclipped counterparts. We trace this failure to the clipped objective. We prove that the clipped objective can fail to correct the student toward the teacher and can instead push clipped and unclipped token probabilities away from its teacher. Our training runs agree with this analysis: inside repetitions, the clipped student places less probability than its teacher on leaving the repetition, and more on continuing it, whereas the unclipped runs stay close to their teachers.