On-Policy Self-Distillation with Sampled Demonstrations Reduces Output Diversity

• URL: http://arxiv.org/abs/2606.26091v1
• Date: Wed, 24 Jun 2026 17:59:02 GMT
• Title: On-Policy Self-Distillation with Sampled Demonstrations Reduces Output Diversity
• Abstract summary: On-policy self-distillation achieves strong pass@1 accuracy by using a single model as both teacher and student.<n>We show that this could come at a hidden cost: rollout diversity decreases and pass@k curves flatten.
• Score: 5.35852583397955
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: On-policy self-distillation achieves strong pass@1 accuracy by using a single model as both teacher and student, with the teacher conditioned on a correct demonstration to provide dense token-level feedback. We show that this could come at a hidden cost: rollout diversity decreases and pass@k curves flatten (i.e., generating more rollouts fails to improve accuracy). We trace this to compounding biases in the design of self-distillation with sampled demonstrations. The teacher scores each student rollout while conditioned on a sampled correct rollout, channeling its feedback through the model's own biases. We theoretically analyze the optimal self-distillation policy and show that it tilts the base distribution by a pointwise conditional mutual information score between the student's rollout and the correct rollout used as context. Unlike the ideal optimal on-policy reinforcement learning (RL), which preserves probability ratios among equally correct rollouts, self-distillation can amplify existing probability gaps, concentrating mass on already-dominant modes. On a controlled graph path-finding task and science question-answering benchmarks, self-distilled models match or exceed RL on average performance but exhibit substantially lower functional and semantic diversity, failing on out-of-distribution settings that require diverse strategies.

Related papers

• Self-OPD: On-Policy Distillation for Flow Matching Models without Teacher [36.07614733558147]
  We introduce textbfSelf-OPD, a teacher-free OPD framework for flow matching models.<n>Self-OPD branches the deterministic next-state prediction into $K$ SDE candidates, rolls them out with the ODE sampler, and compares their rewards against a deterministic self-reference baseline.<n>Experiments on single and mixed reward benchmarks show that Self-OPD outperforms prior RL and OPD methods without task-specific teachers.
  arXiv  Detail & Related papers  (2026-08-27T09:34:23Z)
• Contrastive On-Policy Distillation [36.47487086937924]
  We propose COPD, a contrastive On-policy Distillation framework.<n> COPD directly encourages the student model to learn more concise and efficient reasoning strategies.
  arXiv  Detail & Related papers  (2026-07-21T12:35:04Z)
• A Formula-Driven Survey and Research Agenda for On-Policy Distillation [4.397842507533513]
  This survey studies OPD as a feedback-to-update problem rather than a single loss family.<n>We develop a formula-driven taxonomy from two routes -- direct distributional losses and policy-gradient-style log-ratio updates.
  arXiv  Detail & Related papers  (2026-06-22T03:09:21Z)
• DeltaPrompts: Escaping the Zero-Delta Trap in Multimodal Distillation [49.98710755440242]
  Distillation enables compact Vision-Language Models (VLMs) to obtain strong reasoning capabilities.<n>We reveal a critical inefficiency in this approach: up to 69% of the prompts in standard chart / document reasoning datasets are effectively zero-delta.<n>We propose a staged synthesis pipeline that repurposes existing datasets as seeds, actively targeting student failure modes to produce better prompts.
  arXiv  Detail & Related papers  (2026-05-15T02:04:12Z)
• DiffusionOPD: A Unified Perspective of On-Policy Distillation in Diffusion Models [55.01951088768769]
  DiffusionOPD is a new multi-task training paradigm for diffusion models based on Online Policy Distillation (OPD)<n>We show that DiffusionOPD consistently surpasses both multi-reward RL and cascade RL baselines in training efficiency and final performance.
  arXiv  Detail & Related papers  (2026-05-14T16:49:09Z)
• Multi-Rollout On-Policy Distillation via Peer Successes and Failures [31.731266689209914]
  Large language models are often post-trained with sparse verifier rewards, which indicate whether a sampled trajectory succeeds but provide limited guidance about where reasoning succeeds or fails.<n>On-policy distillation (OPD) offers denser token-level supervision by training on student-generated trajectories.<n>We introduce Multi-Rollout On-Policy Distillation (MOPD), a peer-conditioned distillation framework that uses the student's local rollout group to construct more informative teacher signals.
  arXiv  Detail & Related papers  (2026-05-12T18:57:44Z)
• Preference-Based Self-Distillation: Beyond KL Matching via Reward Regularization [18.027254451537342]
  Existing self-distillation methods largely reduce learning to KL matching toward the context-augmented teacher model.<n>We propose textbfPreference-textbfBased textbfSelf-textbfDistillation (textbfPBSD), which revisits on-policy self-distillation through a reward-regularized perspective.
  arXiv  Detail & Related papers  (2026-05-06T15:31:50Z)
• PAINT: Partial-Solution Adaptive Interpolated Training for Self-Distilled Reasoners [9.324642081509756]
  Large language model (LLM) reasoning requires supervision that is both aligned with the model's own test-time states and informative at the token level.<n>Recent privileged on-policy self-distillation explores a middle ground by scoring student rollouts with the same model under verified solution context.<n>We propose PAINT, which masks the verified solution according to rollout-reference overlap and applies a small energy-space on a sparse set of entropy-mismatch token positions.
  arXiv  Detail & Related papers  (2026-04-29T11:56:07Z)
• Don't Ignore the Tail: Decoupling top-K Probabilities for Efficient Language Model Distillation [50.19746127327559]
  We propose a new tail-aware divergence that decouples the contribution of the teacher model's top-K predicted probabilities from that of lower-probability predictions.<n> Experimental results demonstrate that our modified distillation method yields competitive performance in both pre-training and supervised distillation of decoder models.
  arXiv  Detail & Related papers  (2026-02-24T11:54:06Z)
• Noisy Correspondence Learning with Self-Reinforcing Errors Mitigation [63.180725016463974]
  Cross-modal retrieval relies on well-matched large-scale datasets that are laborious in practice. We introduce a novel noisy correspondence learning framework, namely textbfSelf-textbfReinforcing textbfErrors textbfMitigation (SREM)
  arXiv  Detail & Related papers  (2023-12-27T09:03:43Z)
• Delving into Identify-Emphasize Paradigm for Combating Unknown Bias [52.76758938921129]
  We propose an effective bias-conflicting scoring method (ECS) to boost the identification accuracy. We also propose gradient alignment (GA) to balance the contributions of the mined bias-aligned and bias-conflicting samples. Experiments are conducted on multiple datasets in various settings, demonstrating that the proposed solution can mitigate the impact of unknown biases.
  arXiv  Detail & Related papers  (2023-02-22T14:50:24Z)
• Distributionally Robust Models with Parametric Likelihood Ratios [123.05074253513935]
  Three simple ideas allow us to train models with DRO using a broader class of parametric likelihood ratios. We find that models trained with the resulting parametric adversaries are consistently more robust to subpopulation shifts when compared to other DRO approaches.
  arXiv  Detail & Related papers  (2022-04-13T12:43:12Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.