Showing 1–25 of 666 results for abstract: "on-policy distillation" OR "trajectory distillation" OR "agent distillation" OR "distilling agent"

PivotOPD: Learning to Recover from Pivotal Mistakes in Multi-Turn Agents

Authors: Yinghui He, Yapei Chang, Khushi Bhardwaj, Daniele Molinari, Tugrul Konuk, Jan Kautz, Ali Hatamizadeh

Abstract: On-policy… ▽ More On-policy distillation (OPD) is a promising approach for training language agents, providing dense teacher supervision on student-generated trajectories. However, in multi-turn interaction, an incorrect action changes the states the student encounters later, so errors compound across turns. In preliminary experiments across three Qwen3 models (8B to 235B), we find that more than half of the failed rollouts contain a pivotal mistake, an action that moves the agent farther from completing the task, and this mistake typically occurs early. These pivotal mistakes often remain recoverable: guiding the model for only a few turns after the pivotal turn can restore task success. We therefore propose PivotOPD, an on-policy distillation framework that jointly trains the student to prevent pivotal mistakes and to recover from the states they create. At each pivotal mistake, a teacher model provides a gold action and then names a recovery action at each of the next few turns. Preventive distillation uses the gold action with reverse KL to steer the student away from the pivotal mistake, while recovery distillation uses the recovery actions with forward KL to transfer recovery behaviors that the student rarely samples. Against 13 baselines on ALFWorld, WebShop, and Search-based QA, PivotOPD achieves the strongest average performance for both Qwen3-1.7B and Qwen3-8B students, improving over the strongest baseline on ALFWorld by +5.5% with the 1.7B student. The gains also transfer to another model family on the software engineering domain, where PivotOPD raises the resolve rate of a Nemotron-3.5 student on SWE-Bench Verified by +3.2%. Project page: https://research.nvidia.com/labs/lpr/pivotopd/ △ Less

Submitted 30 September, 2026; originally announced September 2026.

Comments: PivotOPD technical report; Project page: https://research.nvidia.com/labs/lpr/pivotopd/

Less Data, Better Timing: Student-Curriculum Coupling for VLM On-Policy Distillation in Temporal Video Grounding

Authors: Jiacheng Qiu, Yunsoo Kim, Ruichen Xu, Jian Luo, Petar M. Djurić, Sima Mofakham

Abstract: On-policy distillation (OPD) provides dense supervision directly on student-generated trajectories, making it an effective post-training strategy for vision-language models in temporal video grounding (TVG). However, existing pipelines typically construct the training curriculum from a fixed teacher and the initial stu… ▽ More On-policy distillation (OPD) provides dense supervision directly on student-generated trajectories, making it an effective post-training strategy for vision-language models in temporal video grounding (TVG). However, existing pipelines typically construct the training curriculum from a fixed teacher and the initial student state, implicitly assuming that selected examples retain positive supervision value throughout optimization. We show that supervision trustworthiness and supervision necessity are distinct yet coupled: the former concerns target credibility, while the latter varies with the student's current task competence; together, they shape supervision value. Building on this coupled view, we introduce Student-Curriculum Coupling (SCC), a closed-loop framework in which a compact Anchor-Frontier curriculum defines the candidate supervision space and the evolving student dynamically determines its active subset. Supervision can therefore be activated, suspended, or reactivated as competence changes, concentrating teacher computation and optimization on current task-level deficits. Across three TVG benchmarks, SCC achieves a 5.1% relative improvement in mean recall over Video-OPD on its original curriculum, while using 60.0% fewer training examples and reducing training time by 50.4%. Ablations support the complementary roles of capability-structured curriculum design and student-dependent supervision in achieving these gains. Together, these results establish SCC as a data- and compute-efficient framework for TVG post-training, delivering stronger temporal grounding by aligning trustworthy supervision with the student's evolving learning needs. △ Less

Submitted 30 September, 2026; originally announced September 2026.

LatentHarness: Learning Latent Actions for Memory and Reasoning via Counterfactual Policy Distillation

Authors: Xiaoqiang Wang, Suyuchen Wang, Bang Liu

Abstract: …computation, RECALL from a fast-weight memory of input evidence and intermediate reasoning states, or EXIT to emit the next token. We train this policy with counterfactual policy distillation, which branches every action for one step and scores its effect on the emitted token. These gains teach the policy when memory i… ▽ More Long-context reasoning faces two complementary bottlenecks: retaining evidence across long inputs and sustaining computation across many reasoning steps. Existing approaches largely address them separately, with external memory extending access to distant evidence and latent reasoning compressing multi-step computation. We introduce LatentHarness, which unifies memory access and latent reasoning as sequential latent action selection. At each internal step, the model chooses THINK for further computation, RECALL from a fast-weight memory of input evidence and intermediate reasoning states, or EXIT to emit the next token. We train this policy with counterfactual policy distillation, which branches every action for one step and scores its effect on the emitted token. These gains teach the policy when memory is more useful than further reasoning, while gradients through counterfactual recall teach which intermediate states should be retained in memory for future use. Across six general and long-context reasoning benchmarks, LatentHarness at 1.4B improves on the strongest baselines by 2.8% and 10.0% relative, respectively, and runs 5.9x faster than the strongest long-context baseline. △ Less

Submitted 30 September, 2026; originally announced September 2026.

Comments: Work in progress

GFD-OPD: Guidance-Folded On-Policy Distillation of Diffusion Models Across Scales

Authors: Zhenxing Zhang, Jiayan Teng, Wenxu Wu, Zhuoyi Yang, Jiazheng Xu, Wendi Zheng, Jie Tang, Dan Guo, Meng Wang

Abstract: On-policy distillation (OPD) has demonstrated two important capabilities in language models: compressing large teachers into smaller students and merging expert models into a single model. Existing diffusion OPD, however, mostly focus on the latter, with teachers and students sharing the same backbone and scale. We inv… ▽ More On-policy distillation (OPD) has demonstrated two important capabilities in language models: compressing large teachers into smaller students and merging expert models into a single model. Existing diffusion OPD, however, mostly focus on the latter, with teachers and students sharing the same backbone and scale. We investigate large-to-small diffusion opd from large teachers to a small student and find that the standard recipe fails. To find the underlying cause, we propose Fixed-State KL, an effective and fair way to measure the distribution gap between student and teacher during OPD training for diffusion models. We are the first to clarify why large-to-small OPD is challenging for diffusion models: a smaller student struggles to perfectly match the distribution of a larger teacher, while classifier-free guidance can accumulate and amplify the distributional discrepancies between the student's conditional and unconditional branches and those of the teacher. To solve this problem, we propose GFD-OPD, a simple yet effective method that reduces the student-teacher gap while avoiding the error amplification of the CFG composition. Across numerous experiments, GFD outperforms previous baselines in both training efficiency and final performance, achieving state-of-the-art results on all benchmarks. △ Less

Submitted 30 September, 2026; originally announced September 2026.

From Imitation to Reward Discovery: On-Policy Warmup for Agentic RL

Authors: Yitong Qiao, Tiantian He, Lei Liu, Yue Shen, Jian Wang, Jinjie Gu, Zhixuan Chu

Abstract: …rewards can leave early training with little signal for policy improvement. We identify an On-Policy Acceleration Phenomenon: in our main comparisons, RLVR initialized with on-policy… ▽ More Reinforcement learning with a verifiable reward (RLVR) offers a scalable approach to training language-model agents, yet sparse outcome rewards can leave early training with little signal for policy improvement. We identify an On-Policy Acceleration Phenomenon: in our main comparisons, RLVR initialized with on-policy distillation reaches high performance earlier in training and achieves both higher average performance during subsequent RLVR and higher final performance than the alternative baselines. Motivated by this observation, we study On-Policy Warmup (OPW), a teacher-guided stage in which the student trains with teacher supervision on its own interaction trajectories before transitioning to RLVR. Unlike imitation on fixed teacher-generated trajectories, OPW targets states induced by the student's own decisions, including imperfect actions and recovery situations. We provide a theoretical explanation by connecting on-policy reverse-KL distillation to trajectory-level distribution matching. Under a competent teacher and sufficiently small population distillation loss, this connection yields a lower bound on initial verifier success and a corresponding bound on reward-discovery complexity. For group-relative RLVR, we further characterize when increased success probability produces more reward-informative groups. Together, our findings support on-policy distillation as an effective warmup for agentic RLVR and identify initial reward discovery as a mechanism that can contribute to the observed acceleration. △ Less

Submitted 30 September, 2026; originally announced September 2026.

No Task Vector Is an Island: A Comprehensive Study on the Composability of Task Vectors from On-Policy Distillation

Authors: Jingang Zhou, Feiyu Han, Han Zhu, Yuyi Zhou, Ruiyang Zhang, Jian Xu, Sirui Gao, Qingpei Guo, Xu-Yao Zhang

Abstract: Task vectors provide a simple mechanism for composing learned capabilities through model merging. However, the composability of task vectors produced by on-policy distillation (OPD) remains largely unexplored. OPD trains a student using teacher feedback on student-generated trajectories, yielding parameter updates that… ▽ More Task vectors provide a simple mechanism for composing learned capabilities through model merging. However, the composability of task vectors produced by on-policy distillation (OPD) remains largely unexplored. OPD trains a student using teacher feedback on student-generated trajectories, yielding parameter updates that differ from those produced by the teacher model, usually by reinforcement learning (RL). We therefore ask whether OPD task vectors can complement their RL teacher updates and compose effectively across tasks. Across five domains and two model architectures, we find evidence for both forms of composability. Within a task, merging OPD and RL task vectors can outperform both constituent models, even when the OPD student is weaker than its RL teacher. Across tasks, OPD task-vector compositions achieve higher average scores than corresponding RL compositions in seven of eight backbone-merging-rule comparisons. Parameter-space analyses reveal substantial non-collinearity between OPD and RL updates. Experiment in CODE domain on SMOLLM3-3B shows that the combined direction outperforms either constituent direction at the tested global update norm, supporting directional complementarity in this configuration. Across tasks, OPD updates also show lower overlap among the top-10% feed-forward channels ranked by update energy. Together, these results show that weaker standalone performance does not imply weaker task-vector composability. OPD task vectors can complement stronger RL teacher updates and combine effectively across tasks, highlighting composability as a distinct property for understanding and evaluating post-training updates. △ Less

Submitted 30 September, 2026; originally announced September 2026.

Experimental Experience Modeling for Autonomous Research

Authors: Wenda Wei, Yingchen Zhang, Ruqing Zhang, Jiafeng Guo, Daiting Shi, Xueqi Cheng

Abstract: …for making informed experimental decisions by acquiring, reusing, and accumulating experimental experience. EEM extracts decision-relevant records from earlier experimental trajectories, distills them into reusable experience, and organizes them in an experience library. For a new experimental decision, EEM retrieves r… ▽ More Autonomous research agents can generate hypotheses and conduct experiments, but experimentation remains a major source of computational cost. A fundamental challenge is deciding which experiments are worth running, particularly when prior evidence is insufficient to resolve uncertainty. Yet current research agents lack a systematic way to leverage experimental experience when making such decisions. We introduce Experimental Experience Modeling (EEM), a framework for making informed experimental decisions by acquiring, reusing, and accumulating experimental experience. EEM extracts decision-relevant records from earlier experimental trajectories, distills them into reusable experience, and organizes them in an experience library. For a new experimental decision, EEM retrieves relevant historical experience and assesses whether it provides sufficient support for deciding whether a candidate direction warrants further investment. When historical experience is insufficient, EEM conducts a targeted, low-cost pilot experiment to acquire the missing decision-relevant experience on demand. It then combines this newly acquired experience with retrieved historical experience to determine whether the direction warrants full-scale evaluation, which requires substantial resources. The resulting experimental outcomes are further distilled into reusable experience, allowing the library to continually grow through iterative accumulation. Experiments on autonomous research benchmarks show that EEM improves research performance while reducing model interaction overhead, demonstrating the value of reusing accumulated experience and acquiring additional experience only when needed. △ Less

Submitted 30 September, 2026; originally announced September 2026.

Residual Trajectory Distillation for Generative Retrieval

Authors: Weihao Shen, Wei Chen, Fuwei Zhang, Guojun Liu, Qingsong Hua, Wei Lin, Fuzhen Zhuang

Abstract: …collapses distinct quantization behaviors into identical targets and leaves information available during indexing unused in retrieval training. We introduce ResTD, a Residual Trajectory Distillation framework that transfers this discarded indexing information into retrieval training. Treating the frozen RQ indexer as a… ▽ More Generative retrieval has emerged as a general retrieval paradigm, representing items with discrete Semantic IDs (SIDs) and retrieving them through autoregressive identifier generation. When SIDs are constructed with residual quantization (RQ), standard retrieval training supervises only the selected codes and discards the residual trajectories that produce them. The same hard code can nevertheless arise from different preferences over competing codewords, while the residual trajectory also contains information about subsequent quantization decisions. As a result, hard SID supervision collapses distinct quantization behaviors into identical targets and leaves information available during indexing unused in retrieval training. We introduce ResTD, a Residual Trajectory Distillation framework that transfers this discarded indexing information into retrieval training. Treating the frozen RQ indexer as a process teacher, it distills residual-induced codeword preferences into SID-decoding states. This supervision recovers distinctions hidden by hard assignments and allows earlier decoder states to capture information about subsequent quantization decisions before the corresponding SID suffix is generated. In this way, richer information from SID construction is incorporated into retrieval learning while preserving the original retrieval index and inference procedure. Experiments on multilingual e-commerce retrieval show consistent improvements over strong baselines and matched training controls. Controlled comparisons show that residual-derived targets outperform the tested codebook-only soft targets. Representation probes further show that future codebook preferences become more recoverable from earlier decoder states. ResTD can also be readily extended beyond retrieval to generative recommendation. Code is available at: https://github.com/Nevaeh7/iclr2027_ResTD.git. △ Less

Submitted 30 September, 2026; originally announced September 2026.

ReTaCo: Residual-Target Control for On-Policy Distillation

Authors: Zixiang Ni, Zhuo Hu, Renjie Cao, Weijie Ren, Binqin Shi, Weijia Zhang, Shuheng Cao, Zhicheng Shi, Zhenhao Zhang, Haomin Wen, Zhiyuan Hu

Abstract: On-policy distillation (OPD) trains a student on its own generated prefixes with token-level teacher feedback, but transmitting or storing the teacher's full-vocabulary distribution at every token is costly. Entropy-aware OPD (EOPD) adds forward supervision to reverse KL to help the student recover plausible tokens… ▽ More On-policy distillation (OPD) trains a student on its own generated prefixes with token-level teacher feedback, but transmitting or storing the teacher's full-vocabulary distribution at every token is costly. Entropy-aware OPD (EOPD) adds forward supervision to reverse KL to help the student recover plausible tokens it underestimates, using only the teacher's top-$k$ probabilities to limit cost. Because EOPD renormalizes these probabilities, its target assigns no mass to the omitted vocabulary. We prove that the resulting loss keeps pushing the student's top-$k$ mass toward one even after the student matches the teacher's relative probabilities within the top-$k$ set, so the teacher itself is not a stationary point whenever the omitted tokens have positive teacher probability. We propose ReTaCo (Residual-Target Control), which keeps the top-$k$ tokens individually and groups the remaining tokens into one residual symbol, and pairs this forward target with a single-sample estimator whose expectation equals the full-vocabulary reverse KL. With teacher top-$k$ mass $m$, the residual target is $(1-β)(1-m)$ for $β\in[0,1]$: $β=0$ preserves the teacher's mass, and larger $β$ moves more mass onto the top-$k$ tokens without changing their relative probabilities. At a fixed prefix, we prove that the population objective has a unique optimum whose top-$k$ mass lies between $m$ and $m+β(1-m)$ and increases monotonically with $β$; at $β=0$, underestimated top-$k$ tokens still receive non-vanishing recovery gradients. Numerical optimization confirms these predictions, and across three teacher-student pairs, ReTaCo outperforms EOPD on most mathematics and code benchmarks. △ Less

Submitted 30 September, 2026; originally announced September 2026.

OP-CAD: On-Policy Clean-Audio Distillation for Robust Audio-Visual Reasoning

Authors: Xingming Shui, Dapeng Chen, Bowei Liu, Jingqi Tian, Minfu Li, Kun Yi, Jiapeng Hong, Yansong Tang

Abstract: …focusing on question answering under environmental noise and competing speech. The challenge is to resist acoustic interference while preserving useful audio evidence. On-policy distillation provides dense teacher feedback on student-generated responses, but uniform token weighting does not explicitly prioritize posit… ▽ More Omni-modal large language models deployed in real-world environments encounter external noise that can interfere with their perception and understanding of multimodal inputs. We study their robustness in audio-visual understanding, focusing on question answering under environmental noise and competing speech. The challenge is to resist acoustic interference while preserving useful audio evidence. On-policy distillation provides dense teacher feedback on student-generated responses, but uniform token weighting does not explicitly prioritize positions affected by acoustic interference. We introduce OP-CAD (On-Policy Clean-Audio Distillation), a curriculum-based privileged self-distillation framework for robust audio-visual understanding. Training progresses from mild to severe environmental noise and competing speech, with selective token-level supervision at each stage. The student generates responses from corrupted audio-visual input, while a frozen teacher uses clean audio and the verified answer to supervise the same response prefixes. To allocate this supervision, OP-CAD compares teacher predictions under clean, corrupted, and visual-only contexts without revealing the answer. These matched comparisons measure sensitivity to audio removal and corruption; a bounded weighting rule emphasizes positions identified by either signal while retaining supervision throughout the response. OP-CAD outperforms the compared methods across all evaluated noise conditions. Paired analyses further show improved preservation of clean-correct answers under strong interference, with no observed aggregate clean-accuracy penalty. These results demonstrate the value of directing clean-teacher supervision toward acoustically sensitive predictions for robust audio-visual reasoning. △ Less

Submitted 30 September, 2026; originally announced September 2026.

Is Better Teacher Supervision Enough? Unlocking Student-side Learning in Multimodal On-Policy Distillation

Authors: Siyuan Liu, Kanghui Tian, Yue Duan, Yutao He, Shangdong Yang, Jian Zhang, Yinghuan Shi

Abstract: On-policy… ▽ More On-policy distillation (OPD) improves reasoning by providing token-level supervision from a teacher on a student's own trajectories. Existing methods primarily focus on enhancing this teacher-side guidance (e.g., by enriching teacher inputs and refining teacher feedback), yet we find that limited student perception is another critical bottleneck in multimodal OPD. By providing oracle visual facts, the performance of OPD-trained students can still be substantially improved for both weak and strong teachers. To address this bottleneck, we propose S-OPD, a simple multimodal on-policy distillation framework that explicitly strengthens student perceptual learning through two objectives. Specifically, Teacher-calibrated Policy Contrast separates student policies under original and masked images with teacher-based token-level gating, strengthening the student's reliance on visual evidence during reasoning. Policy Agreement aligns student policies under original and noise-perturbed images, further improving perceptual robustness to visual noise. Notably, our method can be seamlessly plugged into existing OPD frameworks, requiring no additional data annotations, model parameters or inference operations. Extensive experiments on eight benchmarks across student scales and distillation paradigms demonstrate consistent performance improvements, with gains of up to 4.25 points on LogicVista. When combined with existing teacher-side supervision methods, our method can yield further gains. Code is available at https://github.com/Sirilaw/S-OPD. △ Less

Submitted 30 September, 2026; originally announced September 2026.

Mitigating the Length-Scaling Tax with Online Distillation

Authors: Xu Wan, Wenyue Xu, Shengjie Zhao, Mingyang Sun

Abstract: …response length on already-solved queries without a commensurate accuracy gain. To mitigate LST, we propose Length Self-Distillation (LSD), which routes solved prompts to on-policy distillation and retains the original RL objective for unsolved prompts. LSD uses an exponential moving average of the online policy as its… ▽ More Length scaling during reinforcement-learning (RL) post-training is often viewed as a sign of improved reasoning ability, especially on difficult problems, but may also make responses to already-solved problems unnecessarily verbose. We quantify this side effect as the length-scaling tax (LST): excess response length on already-solved queries without a commensurate accuracy gain. To mitigate LST, we propose Length Self-Distillation (LSD), which routes solved prompts to on-policy distillation and retains the original RL objective for unsolved prompts. LSD uses an exponential moving average of the online policy as its teacher, requiring no external model. We find that LSD achieves comparable or better performance than RL across multiple variants, while substantially curbing response-length growth on easy queries. LSD reduces LST from 19.0% to -3.7% on single-turn reasoning and from 31.4% to 13.7% on multi-turn agentic tasks, demonstrating that LSD effectively preserves concise response patterns on easy queries while supporting efficient exploration on difficult queries during RL post-training. △ Less

Submitted 29 September, 2026; originally announced September 2026.

Can Terminal Agents Trust Their Own Verification? Diagnosing and Improving Self-Verification

Authors: Yingfeng Luo, Shaowei Wei, Daixin Wang, Dingyang Lin, Kaiyan Chang, Weiqiao Shan, Tong Zheng, Zhiqiang Zhang, Jingbo Zhu, Tong Xiao

Abstract: …SCVD improves \textsc{Pass@1} on TerminalBench2.1 by 9.74--16.85 percentage points over the corresponding base models and by 4.49--8.61 points over the standard full-trajectory… ▽ More Terminal agents rely on self-verification to assess and correct their solutions as they solve tasks through interaction with command-line environments. Yet how trustworthy such self-verification is remains poorly understood. To investigate this question systematically, we introduce a diagnostic framework that identifies the first complete solution in each trajectory, determines whether it is objectively correct, and uses this ground truth to quantify the agent's subsequent verification and recovery behavior. Applying it to ten terminal agents on TerminalBench2.1, we find that verification is nearly universal after a complete candidate is formed, yet only 61.43\% of incorrect candidates are detected and only 49.36\% of detected errors are successfully repaired. These results show that the main weakness in self-verification lies not in initiating verification, but in detecting and repairing errors. Motivated by these findings, we propose Student-Conditioned Verification Distillation (SCVD), which lets the student first produce a candidate solution and distills a stronger teacher's subsequent verification and recovery from the same interaction context. Across three Qwen3.5 backbones, SCVD improves \textsc{Pass@1} on TerminalBench2.1 by 9.74--16.85 percentage points over the corresponding base models and by 4.49--8.61 points over the standard full-trajectory distillation, while avoiding the pronounced out-of-distribution degradation of full-trajectory distillation on SWE-bench Verified. △ Less

Submitted 29 September, 2026; originally announced September 2026.

Distill the Visual Evidence, Not Just the Answer: Cross-World On-Policy Distillation for Vision-Language Models

Authors: Yuanhao Sun, Huawei Ji, Jiaxin Ding, Luoyi Fu, Xinbing Wang

Abstract: …raising the question: how can we ensure the student responds to the visual information that actually determines the answer? To this end, we propose \textbf{Cross-World On-Policy… ▽ More A central goal of vision-language model (VLM) distillation is to transfer both the teacher's language capabilities and its visual understanding. However, existing methods primarily supervise the student's output, leaving visual understanding implicit. Our analysis reveals that a student can match the teacher's answer without relying on the same visual evidence, raising the question: how can we ensure the student responds to the visual information that actually determines the answer? To this end, we propose \textbf{Cross-World On-Policy Distillation (CW-OPD)}, which explicitly supervises the student's response to changes in visual evidence. For each example, CW-OPD constructs two visual worlds that share the question and scene context but differ in answer-critical evidence, yielding different answers. We perform on-policy distillation in both worlds and distill the teacher's cross-world belief transition, encouraging the student to match not only \emph{what} the teacher predicts but also \emph{why} its prediction changes with the evidence. A gradient analysis shows that this term is invariant to errors shared by both worlds and supplies a corrective signal invisible to endpoint matching alone. In this way, CW-OPD makes reliance on the relevant visual evidence an explicit distillation target rather than an implicit consequence of output matching. To diagnose whether a model truly grounds its answers in visual evidence, we introduce CWBench, which measures cross-world consistency via Cross-World Pair Accuracy (CWPA). Experiments on Qwen3.5-4B show that CW-OPD outperforms the strongest baseline by \textbf{1.2} points on average, and the 4B student exceeds DeepSeek-V4.1 (552B) by \textbf{22.4} CWPA points on CWBench. Code is released in https://github.com/baokou-fw2/CWAD. △ Less

Submitted 29 September, 2026; originally announced September 2026.

Understanding Off- vs On-Policy Distillation: A Tale of Distinct Training Objectives

Authors: Qiwei Di, Xuheng Li, Kaixuan Ji, Chenggong Zhang, Heyang Zhao, Quanquan Gu

Abstract: On-policy distillation (OPD) learns from teacher feedback on student-generated responses and has shown promise in reducing forgetting relative to supervised fine-tuning (SFT). However, its benefits and fragility remain incompletely understood. We study sequential distillation from multiple teachers, where the student m… ▽ More On-policy distillation (OPD) learns from teacher feedback on student-generated responses and has shown promise in reducing forgetting relative to supervised fine-tuning (SFT). However, its benefits and fragility remain incompletely understood. We study sequential distillation from multiple teachers, where the student minimizes its average divergence from the teachers. Forward Kullback--Leibler (KL) divergence yields a weighted arithmetic mixture, while reverse KL yields a normalized weighted geometric aggregate. We develop algorithms that learn these targets under off-policy and on-policy feedback, respectively, establishing logarithmic regret bounds in the tabular setting and extending the analysis to function approximation. By analyzing these aggregation targets, we identify mechanisms that help explain both the benefits and fragility of OPD. Relative to forward KL, reverse KL can better retain a confident expert's preferences under uninformative feedback, but is more sensitive to teachers that assign very low probabilities to correct responses. Its token-level conditionals also reveal a dependence on continuation distributions that can favor incorrect prefixes over long horizons. △ Less

Submitted 29 September, 2026; originally announced September 2026.

Comments: 68 pages, 4 figures

On the Off-Policy Teacher in On-Policy Distillation

Authors: Langlin Huang, Hao Liu, Mononito Goswami, Xinyu Li, Prithwith Jana, Nikos Kanakaris, Patrick Blöbaum, Purak Jain

Abstract: On-policy… ▽ More On-policy distillation (OPD) has recently emerged as a promising post-training paradigm in which the student learns from trajectories generated by its own policy under dense teacher supervision. However, OPD introduces a fundamental asymmetry: although the sampled trajectories are on-policy for the student, they are off-policy for the teacher. The teacher is typically optimized to continue from prefixes generated by its own policy, but during OPD it must instead supervise prefixes generated by the student. Empirically, we find that its continuation performance degrades as these prefixes grow longer. To address this issue, we propose Student-COnditioned Updates of the Teacher (SCOUT), a co-training framework that adapts the teacher to student-generated prefixes. Alongside standard OPD updates, SCOUT periodically optimizes the teacher's conditional ability using reinforcement learning with verifiable rewards, where the teacher generates continuations from student prefixes and learns from outcome rewards. Controlled experiments show that SCOUT improves the teacher's ability to continue from student-generated prefixes, supporting the intended mechanism of student-conditioned teacher adaptation. Across multiple teacher--student configurations, model scales, and reasoning domains, SCOUT also consistently improves the effectiveness of on-policy distillation. △ Less

Submitted 29 September, 2026; originally announced September 2026.

HelixWorld: A Real-time Interactive Audio-Visual World Model

Authors: Lei Ke, Jiahao Pan, Zeyue Tian, Jiaming Wang, Haoyuan Huang, Kam Man Wu, Pengjun Fang, Hongyu Liu, Chenyang Qi, Lin Wang, Ruibin Yuan, Weijia Chen, Fangneng Zhan, Qifeng Chen, Wei Xue, Yike Guo

Abstract: …conditioned on 6-DoF camera trajectories and user actions. To enable low-latency causal interaction, we distill the teacher into a few-step streaming student via an online trajectory distillation loss, sustaining drift-free joint audio-visual rollouts at 24 FPS on a single GPU. Furthermore, we formalize spatial-acousti… ▽ More World simulation is inherently multisensory, demanding synchronized visual and acoustic dynamics in real time. Yet prevailing interactive world models remain strictly silent, focusing exclusively on visual rendering and control while overlooking the acoustic dimension. We present HelixWorld, a real-time interactive audio-visual world model where visual scenes and camera-grounded spatial stereo sound co-evolve natively under user interaction. We curate a high-fidelity spatial audio-visual dataset with true stereo acoustics and metric camera poses, upon which we pre-train a bidirectional teacher conditioned on 6-DoF camera trajectories and user actions. To enable low-latency causal interaction, we distill the teacher into a few-step streaming student via an online trajectory distillation loss, sustaining drift-free joint audio-visual rollouts at 24 FPS on a single GPU. Furthermore, we formalize spatial-acoustic consistency and introduce HelixBench to evaluate whether synthesized sound fields faithfully track dynamic viewpoint motion. Extensive experiments demonstrate that HelixWorld matches state-of-the-art silent world models in visual fidelity and responsiveness, while significantly surpassing existing baselines in camera-aligned spatial-acoustic immersion. △ Less

Submitted 29 September, 2026; originally announced September 2026.

Character Training for Risk-Averse Agents

Authors: Arav Dhoot, Punya Syon Pandey, Jamie Johnson, Daniel Tan, Elliott Thornley, David Demitri Africa

Abstract: …risk preferences. To do this, we construct a model constitution describing constant absolute risk aversion (CARA) over an agent's resources and instill it through on-policy distillation. Despite never seeing the benchmark's decision format during training, character-trained models are competitive with baselines… ▽ More Risk aversion in resources could prevent misaligned AI agents from causing catastrophic harm. Misaligned but risk-averse agents would tend to favor safer strategies like making deals with humans over riskier strategies like rebelling. We train agents to be risk averse through character training, finding that persona traits provide a robust mechanism for instilling risk preferences. To do this, we construct a model constitution describing constant absolute risk aversion (CARA) over an agent's resources and instill it through on-policy distillation. Despite never seeing the benchmark's decision format during training, character-trained models are competitive with baselines trained directly on it, and generalise better than them out of distribution on two of our four models. We also modulate different aspects of the constitution, finding that token budget and model choice are the most influential aspect of character training to instill risk aversion. We conclude from these results that character training is a promising and scalable way to instil broad dispositions, which we can use to our advantage in mitigating risk from misaligned AI agents. △ Less

Submitted 29 September, 2026; originally announced September 2026.

VISTA: Internalizing Collective Visual Experience via On-Policy Distillation for Active Multimodal Agents

Authors: Zheng Jiang, Houde Qian, Yiming Chen, Ling Li, Chaoyang Li, Yueqi Li, Yuxuan Liu, Lifeng Sun

Abstract: …use the group to estimate scalar advantages, leaving complementary visual discoveries underused. We introduce VISTA, which internalizes collective visual experience through on-policy distillation by turning observations from same-input rollouts into shared supervision. Collective visual experience distillation (CVED) o… ▽ More Active multimodal agents use visual tools to acquire task-relevant evidence while reasoning. Although reinforcement learning samples multiple interaction trajectories per input, outcome-based objectives primarily use the group to estimate scalar advantages, leaving complementary visual discoveries underused. We introduce VISTA, which internalizes collective visual experience through on-policy distillation by turning observations from same-input rollouts into shared supervision. Collective visual experience distillation (CVED) organizes these observations with their interaction context and aligns them with individual decisions, while heterogeneity-aware policy improvement (HAPI) reinforces successful trajectories and provides experience-guided distillation for unsuccessful attempts. An experience-conditioned teacher evaluates the student's sampled response prefixes, allowing discoveries from one trajectory to guide learning in another without replacing the student's original history or generating new target trajectories. The trained agent retains its visual tools and acts using its own interaction history. VISTA achieves the strongest average performance among the evaluated active multimodal agents of comparable size and consistently outperforms same-backbone training baselines across fine-grained perception and general reasoning tasks, demonstrating the value of collective experience for active multimodal learning. △ Less

Submitted 29 September, 2026; originally announced September 2026.

Dr. OPD: Learning What to Follow for Optimal On-Policy Distillation of Large Language Models

Authors: Zhenyu Wang, Tianze Wang, Linjun Zhang, Yifan Hu

Abstract: On-policy distillation (OPD) trains a student on its own generated responses using dense, token-level supervision from a stronger teacher. Vanilla OPD treats all teacher signals equally, assuming that the teacher's supervision is equally important for every token. However, teacher signals at different tokens may ha… ▽ More On-policy distillation (OPD) trains a student on its own generated responses using dense, token-level supervision from a stronger teacher. Vanilla OPD treats all teacher signals equally, assuming that the teacher's supervision is equally important for every token. However, teacher signals at different tokens may have very different effects on the student's performance: some correct important reasoning errors, while others have little effect on the final answer. Motivated by this observation, we introduce Dr. OPD (OPD Done Right), which defines the optimal weighted OPD to maximize the student's performance. We formulate Dr. OPD as a bilevel optimization problem in which the student learns from weighted teacher supervision, while the weights are selected to maximize the expected reward of the resulting student. To solve Dr. OPD, we develop an efficient iterative solver that updates the token weights and student policy alternatively. At each round, it updates weights in closed form and then takes one gradient step on the resulting weighted OPD objective. Under regularity conditions, we show that this weighted update achieves a higher expected reward than a vanilla OPD update. Empirically, across strong-to-weak and same-size distillation on math and code, Dr. OPD consistently outperforms all evaluated baselines. In particular, in the strong-to-weak distillation setting, Dr. OPD improves average math performance by $9.7$ points over vanilla OPD, and enables the smaller student to surpass its larger teacher. △ Less

Submitted 29 September, 2026; originally announced September 2026.

Guide, Then Let Go: Gap-Adaptive Teacher Scheduling for Sparse-Reward Agentic RL

Authors: Youling Huang, Tiankuo Xu, Jiaji Liu, Tong Zheng, Shuo Zhou, Shaotong Qi, Junchi Yao, Shiyang Liu, Hao Xu, Pengcheng Xu, Bo Huang, Hongyi Fu, Lin Lin

Abstract: …to a severe cold-start problem, as early-stage policies often fail to solve sampled tasks, leaving little useful reward signal for learning. To mitigate this problem, we use on-policy distillation (OPD) to provide token-level guidance on the student's own rollouts. We find that the benefit of this guidance depends… ▽ More Reinforcement learning for long-horizon agents typically relies on sparse outcome-based rewards. This leads to a severe cold-start problem, as early-stage policies often fail to solve sampled tasks, leaving little useful reward signal for learning. To mitigate this problem, we use on-policy distillation (OPD) to provide token-level guidance on the student's own rollouts. We find that the benefit of this guidance depends on the performance gap between the teacher and the student. When the teacher substantially outperforms the student, distillation helps guide the student through the early training stage where outcome rewards provide little learning signal. As the gap narrows and eventually reverses, however, continued distillation becomes less beneficial and may hinder further improvement. Motivated by this observation, we propose Gap-Adaptive Teacher Scheduling (GATS), which augments the student's RL objective with an OPD term whose weight adapts to the teacher-student performance gap. Specifically, GATS gradually reduces teacher guidance as the student approaches the teacher's reference performance and withdraws it once that reference is reached. This enables GATS to leverage task-trained teachers smaller than the student, since teacher guidance is primarily needed during early training. Across ALFWorld, WebShop, and ScienceWorld with three Qwen2.5 teacher-student configurations, GATS achieves the highest average success rate among the compared methods in all three configurations, improving over reward-only GRPO by 4.37%-11.87% under matched student rollout budgets. Code is available at https://github.com/Ricardo-H/guide-then-let-go. △ Less

Submitted 29 September, 2026; originally announced September 2026.

FlowMap-OPD: Rollout--Kernel Separation for On-Policy Distillation of Few-Step Flow-Map Generators

Authors: Zhiqi Li, Bo Zhu

Abstract: Few-step flow-map generators, including MeanFlow and consistency models, enable efficient sampling through long-range transport, yet their on-policy distillation remains underexplored. We introduce FlowMap-OPD, an on-policy distillation fra… ▽ More Few-step flow-map generators, including MeanFlow and consistency models, enable efficient sampling through long-range transport, yet their on-policy distillation remains underexplored. We introduce FlowMap-OPD, an on-policy distillation framework that separates student-state acquisition from teacher--student distribution comparison. A formulation based on state marginals establishes this separation, while flow--velocity consistency connects local supervision to the deployed long-range map. Within this framework, we develop flow-map, induced-velocity, and instantaneous-velocity distribution supervision, each paired with a separately specified native flow-map rollout. Cross-capacity ImageNet experiments across three teacher rewards identify instantaneous-velocity distribution supervision with independently tunable student consistency as the most effective choice. In text-to-image experiments, FlowMap-OPD demonstrates strong multi-specialist consolidation capabilities and surpasses multi-reward Flow-Map GRPO in task performance and convergence speed. △ Less

Submitted 29 September, 2026; originally announced September 2026.

Comments: 38 pages, 18 figures

PolyOCR-Venus: Unified OCR Foundation Models for Text-Centric Visual Intelligence

Authors: GuangJian Team, Kaili Huang, Yongshuo Zhang, Bingtao Fu, Changjiang Jiang, Chenfan Qu, Chenfeng Zhang, Fangming Cui, Gaoyang Zhang, Jiangwei Xie, Jianshu Li, Jing Huang, Jingwen Bai, Mingqi Fang, Tao Fang, Weihong Zhang, Wenbo Du, Xiongfei Bai, Xuekang Zhu, Yinan Xia, Zhenming Wang, Jian Liu, Jingjing Liu, Xiang Qi, Weiqiang Wang

Abstract: …resources into quality-verified OCR supervision. We introduce Competence-Guided Policy Optimization, which combines verifier-based Group Relative Policy Optimization with on-policy distillation through sample-wise routing based on teacher reliability and the teacher--student competence gap. We also introduce OCRBench v… ▽ More Optical Character Recognition (OCR) is evolving from plain-text transcription toward general visual intelligence, requiring models to recognize, localize, and reason over textual information in complex visual environments. However, existing OCR systems often excel at only some tasks and struggle to balance recognition, parsing, and reasoning across scenarios. In this report, we present PolyOCR, a family of unified OCR foundation models of varying scales. PolyOCR combines a shared instruction-following framework with a large-scale data engine that converts heterogeneous visual resources into quality-verified OCR supervision. We introduce Competence-Guided Policy Optimization, which combines verifier-based Group Relative Policy Optimization with on-policy distillation through sample-wise routing based on teacher reliability and the teacher--student competence gap. We also introduce OCRBench v2.1, our revision of OCRBench v2 with manually verified annotation corrections and task-aligned scoring metrics. Extensive experiments across OCRBench v2.1, CC-OCR, in-house KIE Benchmark, OmniDocBench v1.6 and MDPBench demonstrate that PolyOCR achieves state-of-the-art or highly competitive performance. △ Less

Submitted 29 September, 2026; originally announced September 2026.

Comments: Technical Report

Graph-Conditioned On-Policy Agent Distillation from Off-the-Shelf Teachers

Authors: Xiaohan Yi, Wen Luo, Yani Huang, Junfeng Zhan, Asher Qin, Peilin Zhao, Xi Xiao

Abstract: On-policy… ▽ More On-policy distillation (OPD) trains compact language agents with teacher feedback on student-generated trajectories. In multi-turn tasks, compounding errors can move students beyond the teacher's effective supervision. We introduce Graph-Conditioned On-Policy Agent Distillation (GC-OPD), which enriches an off-the-shelf teacher's scoring context with execution evidence. A graph indexes repeated teacher executions by shared states while preserving complete successful and failed histories. After each student episode, GC-OPD retrieves current-state references or historical alternatives and combines them with student hindsight to score the original thought-action tokens. Using the same original teachers, GC-OPD improves mean success over vanilla OPD from 24.70% to 48.78% on ScienceWorld (4B student), from 53.36% to 85.26% on ALFWorld Unseen, and from 29.10% to 37.65% on WebShop. At matched student sizes, it also achieves higher mean success than every evaluated OPD baseline using GRPO-trained teachers on ScienceWorld and ALFWorld; the strongest such ScienceWorld 4B baseline reaches 46.66%. GC-OPD requires no task-specific teacher optimization. △ Less

Submitted 29 September, 2026; originally announced September 2026.

Comments: 18 pages, 3 figures

From Dissonance to Orchestration: Teacher Intervention in On-Policy Distillation

Authors: Yuhao Wang, Ruiyang Ren, Yinan Zhang, Ruiqing Zhang, Jing Liu, Chunyan Miao

Abstract: On-policy distillation (OPD) trains a student on its own reasoning trajectories using feedback from a stronger teacher. Teacher interventions can improve these trajectories, but also change the distribution on which the student learns. Our controlled studies show that rollout quality alone is an incomplete criterion fo… ▽ More On-policy distillation (OPD) trains a student on its own reasoning trajectories using feedback from a stronger teacher. Teacher interventions can improve these trajectories, but also change the distribution on which the student learns. Our controlled studies show that rollout quality alone is an incomplete criterion for allocating teacher guidance. Deeper intervention yields diminishing gains in rollout accuracy while increasing off-policy load. In a training probe with a restricted rollout horizon, peak student accuracy and performance retention favor different intervention strengths. The preferred intervention depth and placement also vary across benchmarks. These findings motivate MAESTRO, which uses local policy disagreement to jointly adapt when the teacher takes over and how long it generates. Its {policy disagreement score} combines teacher-weighted candidate coverage with local distribution similarity and is aggregated within reasoning paragraphs. Across eight mathematical reasoning benchmarks, MAESTRO achieves the highest macro-average accuracy among the compared methods for both 0.6B and 1.7B Qwen3 students, with the 1.7B student leading on every benchmark. MAESTRO also reduces average training response length by 67.3\% relative to standard OPD. The code is available at https://github.com/yhao-wang/MAESTRO. △ Less

Submitted 29 September, 2026; originally announced September 2026.