Uni-OPD: Unifying On-Policy Distillation with a Dual-Perspective Recipe

• URL: http://arxiv.org/abs/2605.03677v1
• Date: Tue, 05 May 2026 12:15:21 GMT
• Title: Uni-OPD: Unifying On-Policy Distillation with a Dual-Perspective Recipe
• Abstract summary: We propose Uni-OPD, a unified OPD framework that generalizes across Large Language Models (LLMs) and Multimodal Large Language Models (MLLMs)<n>Specifically, from the student's perspective, we adopt two data balancing strategies to promote exploration of informative student-generated states during training.<n>We develop an outcome-guided margin calibration mechanism to restore order consistency between correct and incorrect trajectories.
• Score: 53.40076304466524
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: On-policy distillation (OPD) has recently emerged as an effective post-training paradigm for consolidating the capabilities of specialized expert models into a single student model. Despite its empirical success, the conditions under which OPD yields reliable improvement remain poorly understood. In this work, we identify two fundamental bottlenecks that limit effective OPD: insufficient exploration of informative states and unreliable teacher supervision for student rollouts. Building on this insight, we propose Uni-OPD, a unified OPD framework that generalizes across Large Language Models (LLMs) and Multimodal Large Language Models (MLLMs), centered on a dual-perspective optimization strategy. Specifically, from the student's perspective, we adopt two data balancing strategies to promote exploration of informative student-generated states during training. From the teacher's perspective, we show that reliable supervision hinges on whether aggregated token-level guidance remains order-consistent with the outcome reward. To this end, we develop an outcome-guided margin calibration mechanism to restore order consistency between correct and incorrect trajectories. We conduct extensive experiments on 5 domains and 16 benchmarks covering diverse settings, including single-teacher and multi-teacher distillation across LLMs and MLLMs, strong-to-weak distillation, and cross-modal distillation. Our results verify the effectiveness and versatility of Uni-OPD and provide practical insights into reliable OPD.

Related papers

• 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)
• DOPD: Dual On-policy Distillation [59.65986569617147]
  On-policy distillation offers superior capacity transfer by supervising student-sampled trajectories with dense token-level signals.<n>We propose DOPD, an advantage-aware dual distillation paradigm that dynamically routes token-level supervision between privileged teacher and privileged student policies.<n>Experiments on both large language model (LLM) and vision-language model (VLM) settings demonstrate that DOPD consistently outperforms Vanilla OPD.
  arXiv  Detail & Related papers  (2026-06-29T17:55:53Z)
• PEARL: Training Socratic Tutors with Pedagogically Aligned Reinforcement Learning [18.478727410749205]
  Large Language Models (LLMs) have shown promise as educational tutors, yet effective tutoring requires more than solving problems.<n>We propose PEARL, a pedagogically aligned reinforcement learning framework for training Socratic tutoring agents.
  arXiv  Detail & Related papers  (2026-05-28T08:25:08Z)
• 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)
• Learning beyond Teacher: Generalized On-Policy Distillation with Reward Extrapolation [57.524909883706556]
  On-policy distillation (OPD) has demonstrated strong empirical gains in improving student performance.<n>This work introduces a flexible reference model and a reward scaling factor that controls the relative weight of the reward term against the KL regularization.<n>In particular, in the setting where we merge the knowledge from different domain experts, obtained by applying domain-specific RL to the same student model, ExOPD enables the student to even surpass the teacher's performance boundary.
  arXiv  Detail & Related papers  (2026-02-12T16:14:29Z)
• From Reasoning LLMs to BERT: A Two-Stage Distillation Framework for Search Relevance [20.096802351171377]
  e-commerce search systems face strict latency requirements that prevent the direct application of Large Language Models.<n>We propose a two-stage reasoning distillation framework to transfer reasoning capabilities from a powerful teacher LLM to a lightweight, deployment-friendly student model.<n>Our framework achieves significant improvements across multiple metrics, validating its effectiveness and practical value.
  arXiv  Detail & Related papers  (2025-10-13T06:46:43Z)
• AdaSwitch: Adaptive Switching Generation for Knowledge Distillation [58.647880811071495]
  Small language models (SLMs) are crucial for applications with strict latency and computational constraints.<n>We propose AdaSwitch, a novel approach that combines on-policy and off-policy generation at the token level.<n>AdaSwitch consistently improves accuracy, offering a practical and effective method for distilling SLMs with acceptable additional overhead.
  arXiv  Detail & Related papers  (2025-10-09T06:38:37Z)
• DistiLLM-2: A Contrastive Approach Boosts the Distillation of LLMs [58.4911494598431]
  DistiLLM-2 is a contrastive approach that simultaneously increases the likelihood of teacher responses and decreases that of student responses.<n>Our experiments show that DistiLLM-2 not only builds high-performing student models across a wide range of tasks, but also supports diverse applications.
  arXiv  Detail & Related papers  (2025-03-10T08:51:32Z)
• Robust Training of Federated Models with Extremely Label Deficiency [84.00832527512148]
  Federated semi-supervised learning (FSSL) has emerged as a powerful paradigm for collaboratively training machine learning models using distributed data with label deficiency. We propose a novel twin-model paradigm, called Twin-sight, designed to enhance mutual guidance by providing insights from different perspectives of labeled and unlabeled data. Our comprehensive experiments on four benchmark datasets provide substantial evidence that Twin-sight can significantly outperform state-of-the-art methods across various experimental settings.
  arXiv  Detail & Related papers  (2024-02-22T10:19:34Z)
• Self-Supervised Monocular Depth Estimation with Self-Reference Distillation and Disparity Offset Refinement [15.012694052674899]
  We propose two novel ideas to improve self-supervised monocular depth estimation. We use a parameter-optimized model as the teacher updated as the training epochs to provide additional supervision. We leverage the contextual consistency between high-scale and low-scale features to obtain multiscale disparity offsets.
  arXiv  Detail & Related papers  (2023-02-20T06:28:52Z)

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.