Skip to content

Recovering General Capabilities via Uncertainty-Calibrated Multi-Teacher On-Policy Distillation

Aug 2026 · 0 citations · 61 references
Computer Science

TL;DR

The proposed Uncertainty-Calibrated MOPD (UCMOPD), which addresses limitations of standard MOPD through two complementary mechanisms, improves the general-capability average over standard MOPD by $4.48\% and $7.86\% while maintaining vertical-domain performance.

Abstract

Specializing large language models to vertical domains improves domain-specific behavior but often degrades general capabilities. We study this trade-off in Multi-Teacher On-Policy Distillation (MOPD), where a specialized model learns from domain and general teachers on its own sampled trajectories. Standard MOPD faces two limitations: ordinary on-policy sampling rarely exposes tokens with large positive teacher--student advantages, and advantage sign alone does not establish whether the proposed update direction is reliable. We propose Uncertainty-Calibrated MOPD (UCMOPD), which addresses these limitations through two complementary mechanisms. Golden-Gain Enhancement combines higher-temperature exploration with a standard-temperature anchor and retains trajectories whose positive learning signal matches or exceeds the prompt-specific anchor. Teacher-Endorsement Filtering then uses centered log-likelihood (CLL) to estimate each retained token's plausibility relative to the teacher's uncertainty and probabilistically preserves updates whose directions are supported by that endorsement. Across role-playing and medical-domain specialization, UCMOPD improves the general-capability average over standard MOPD by $4.48\%$ and $7.86\%$, respectively, while maintaining vertical-domain performance. Component ablations and diagnostic analyses support the intended roles of the two mechanisms: exposing and selecting stronger positive signals at the trajectory level and validating update directions through teacher endorsement at the token level.

View source

Similar papers

#artificial intelligence Preprint Aug 2026

Beyond Teacher Likelihood: Group-Calibrated On-Policy Distillation for Long-Context Reasoning

Group-relative residual calibration can incorporate verifier outcomes without discarding dense token-level guidance, and demonstrates that group-relative residual calibration can incorporate verifier outcomes without discarding dense token-level guidance.

Zhu Zhang, Ji-Xun Wang, Xiao-An Xu et al. · 3 citations
#artificial intelligence Preprint Sep 2026

Teacher Should Think Ahead: Adaptive Continuations for Reliable On-Policy Distillation

This work identifies Teacher Uncertainty Contraction (TUC), a systematic phenomenon whereby the teacher's predictive uncertainty decreases as it continues from a student-generated prefix, and proposes Adaptive-Continuations On-Policy Distillation (AC-OPD), which augments informative states along student rollouts with t...

Jin-Gang Zhou, Yu-Yi Zhou, Hai-Yang Guo et al. · 3 citations
#machine learning Preprint Sep 2026

Distill What You Trust: Reliability-Aware Multi-Teacher On-Policy Distillation

TrustMOPD replaces example-level teacher selection with label-free, token-level supervision allocation and closes the overall-score gap between the initial student and oracle-routed teachers when trained on single- and multi-teacher on-policy distillation.

Jie Sun, Mao Zheng, Ming-Yang Song et al. · 6 citations
#machine learning Preprint Sep 2026

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

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, and effectively combines verifier-driven policy optimization with teacher guidance reg...

Youngrok Park, Sangmin Bae, Hojung Jung et al. · 3 citations
#artificial intelligence Preprint Oct 2026

ResOPD: Tail Residualization for Sparse On-Policy Distillation

ResOPD is established as an efficient, plug-and-play variance reduction primitive for sparse on-policy distillation, which provides unbiased full-vocabulary reverse KL gradient estimation under on-policy sampling, with substantial variance reduction under the same sparse payload in the evaluated settings.

Peng-Hui Yang, Long Xing, Xuan-Lang Dai et al. · 0 citations
#machine learning Preprint Oct 2026

From Gradients to Capabilities: Understanding Multi-Teacher On-Policy Distillation

Multi-teacher on-policy distillation (MOPD) aims to combine the strengths of RL-trained teachers in a single student, but how teacher signals affect parameter changes remains underexplored. We study Qwen3-1.7B with four domain teachers trained with RL from the same initialization as the student, comparing gradients, op...

Si-Qi Zhu, Suo-Zhi Huang, Kai-Xuan Zhang et al. · 3 citations

Related blog posts

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.