Eliciting Weak-to-Strong Generalization with On-Policy Reverse Distillation

Hugging Face Daily Papers Papers

Summary

The paper introduces On-Policy Reverse Distillation (OPRD), a method that enables stronger AI models to exceed weaker supervisors by amplifying verifier-supported policy gradients along the teacher's shift direction, achieving higher performance with fewer updates in distillation scenarios.

Weak-to-strong generalization asks whether stronger models can learn from weaker supervisors and surpass them. This question is particularly important for successive model generations and multi-domain consolidation, where repeating frontier-scale post-training from scratch can be prohibitively expensive. Yet conventional distillation treats the weak teacher as an optimization target, potentially imposing its capacity ceiling on the student. We introduce On-Policy Reverse Distillation (OPRD), which evaluates the teacher's policy shift relative to its reference policy on student rollouts and amplifies the component of the student's verifier-driven policy gradient along that direction. By rescaling only verifier-supported updates, OPRD preserves the stationary points of policy optimization while accelerating learning beyond the teacher. In both successive model transfer and multi-teacher distillation, OPRD achieves higher performance with fewer student updates than existing RL and distillation approaches. Response-style analysis shows that OPRD students remain closer to models trained with verifier-based RL alone than to their weak teachers, suggesting that teacher guidance accelerates rather than redirects the student's own optimization. Results in conventional strong-to-weak distillation further demonstrate that OPRD effectively combines verifier-driven policy optimization with teacher guidance regardless of capacity ordering.
Original Article
View Cached Full Text

Cached at: 09/09/26, 04:28 AM

Paper page - Eliciting Weak-to-Strong Generalization with On-Policy Reverse Distillation

Source: https://huggingface.co/papers/2609.08798

Abstract

On-Policy Reverse Distillation enables stronger models to exceed weak supervisors by amplifying verifier-supported policy gradients along the teacher’s shift direction, accelerating optimization without imposing capacity limits.

Weak-to-strong generalizationasks whether stronger models can learn from weaker supervisors and surpass them. This question is particularly important for successive model generations and multi-domain consolidation, where repeating frontier-scale post-training from scratch can be prohibitively expensive. Yet conventional distillation treats the weak teacher as an optimization target, potentially imposing its capacity ceiling on the student. We introduceOn-Policy Reverse Distillation(OPRD), which evaluates the teacher’s policy shift relative to its reference policy on student rollouts and amplifies the component of the student’s verifier-drivenpolicy gradientalong that direction. By rescaling only verifier-supported updates, OPRD preserves the stationary points of policy optimization while accelerating learning beyond the teacher. In both successive model transfer andmulti-teacher distillation, OPRD achieves higher performance with fewer student updates than existing RL and distillation approaches. Response-style analysis shows that OPRD students remain closer to models trained with verifier-based RL alone than to their weak teachers, suggesting that teacher guidance accelerates rather than redirects the student’s own optimization. Results in conventional strong-to-weak distillation further demonstrate that OPRD effectively combinesverifier-driven policy optimizationwith teacher guidance regardless of capacity ordering.

View arXiv pageView PDFGitHub3Add to collection

Get this paper in your agent:

hf papers read 2609\.08798

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2609.08798 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2609.08798 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2609.08798 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

Weak-to-Strong On-Policy Distillation

arXiv cs.LG

Introduces Weak-to-Strong On-Policy Distillation (W2S-OPD), a framework that improves a strong language model by distilling from multiple weaker models using contrast pairs in logit space, consistently outperforming standard on-policy distillation on math and code benchmarks.

On-Policy Delta Distillation

Hugging Face Daily Papers

The paper introduces On-Policy Delta Distillation (OPD^2), a new distillation reward called the delta signal that captures the difference between a teacher model and its base model before reasoning tuning, providing a more direct signal for transferring reasoning capabilities. Experiments across math, science, and code benchmarks show OPD^2 consistently outperforms conventional on-policy distillation.