On-Policy Distillation

On-policy distillation (OPD) trains an LLM on states induced by the current or recent student policy: the student generates complete or partial rollouts, a teacher or self-teacher scores the resulting tokens under their generated contexts, and dense log-probability, logit, or distributional signals are converted into post-training updates. This survey studies OPD as a feedback-to-update problem rather than a single loss family. We develop a formula-driven taxonomy from two routes -- direct distributional losses and policy-gradient-style log-ratio updates -- and use it to organize core methods, verifier- or outcome-guided hybrids, industrial reports, framework implementations, failure modes, and stabilization recipes under explicit evidence boundaries. The taxonomy shows that OPD effectiveness depends not only on KL direction or teacher access, but also on state compatibility, support construction, temporal credit, vocabulary-level probability routing, gates and weights, and regularization. We further separate two mechanisms often conflated in sampled-token OPD stability discussions. Temporal credit asks how teacher-student log-ratio returns should weight sampled actions across a rollout; vocabulary routing asks where probability mass should move when negative feedback suppresses a sampled token. This distinction yields bias boundaries for immediate, return-to-go, discounted, and baseline-corrected estimators, motivates GAE-OPD as a value-based hypothesis for log-ratio returns, and motivates Counterfactual Routed OPD (CR-OPD) for routing probability mass toward teacher-supported, student-reachable alternatives. We close by mapping actionability diagnostics, failure mechanisms, case studies, open problems, and a reporting checklist onto the same feedback-to-update variables.

On-Policy Distillation (OPD) improves the learning efficiency of standard reinforcement learning through dense, token-level supervision from teachers. In the standard KL objective of OPD, token-level losses are uniformly averaged, implying equal weights for all tokens. However, we discover that not all tokens are created equal: as student rollouts grow longer, they deviate further from the teacher's distribution, leading to degraded supervision quality at later positions. As a result, OPD using only the first 30% of tokens can perform comparably to using all tokens, whereas OPD using only the last 30% of tokens barely learns anything. In this work, we provide a principled understanding of this issue through the lens of constrained optimization. Based on these insights, we derive Importance-Weighted On-Policy Distillation (IW-OPD), in which the weight assigned to each token depends on the accumulated discrepancy between the student's and teacher's distributions, naturally upweighting earlier tokens and downweighting later ones with larger deviations. We show that IW-OPD converges significantly faster than OPD, with better learning efficiency, and achieves better final performance than standard OPD in both same-size and cross-scale settings, improving performance up to 6.9 points on AIME-2025.

On-policy distillation (OPD) improves reasoning models by applying dense teacher supervision on student-sampled trajectories. However, scaling OPD to long-horizon mathematical reasoning exposes a reliability and efficiency problem: standard OPD assigns every sampled candidate the same long rollout budget, even though some trajectories may quickly become weakly aligned with the teacher and provide less useful supervision. Prior analyses suggest that successful OPD depends on local teacher-student compatibility, which can be measured by top-k overlap on student-visited prefixes. When this overlap is low, continuing to generate or train on long suffixes may waste computation and introduce noisy learning signal. To address this, we introduce Prefix-Guided On-Policy Distillation (PG-OPD), a simple rollout-allocation framework that uses fixed-length prefixes to estimate trajectory value before expensive long-horizon generation. PG-OPD first decodes every sampled candidate to the same prefix length, computes teacher-student top-k overlap within an early probe window of that prefix, and selectively continues high-overlap candidates to a fixed long length. Low-overlap candidates stop at the fixed prefix, avoiding unnecessary suffix generation. Across diverse teacher-student combinations on AMC, AIME, and HMMT benchmarks, PG-OPD improves average accuracy by up to 4.80 points while reducing training time by up to 2.46x. These results suggest that prefix-level compatibility provides a practical signal for directing OPD computation toward trajectories that remain learnable from the teacher.

On-policy distillation (OPD) improves student models by training them on trajectories induced by their own policy, making it a promising approach for mitigating exposure bias in agent training. However, most OPD studies focus on single-turn settings, while realistic LLM agents interact with environments over multiple turns. In this regime, early errors can alter future observations and compound across the trajectory, and standard dense token-level OPD becomes brittle, as it may over-penalize semantically valid alternatives, reinforce local degeneracies such as repeated actions, and propagate unreliable teacher supervision on off-distribution histories. We propose SAGE-OPD, a verifier-free selective intervention framework specifically designed for multi-turn OPD. Instead of applying teacher supervision uniformly across all turns, SAGE-OPD first observes environment feedback and uses teacher judgment to decide whether each student response should be skipped or intervened on. To further address compounding errors, SAGE-OPD weights token-level distillation by teacher confidence, reducing the influence of uncertain teacher distributions on corrupted or ambiguous histories. Finally, SAGE-OPD applies loss normalization to preserve the overall loss scale of standard OPD while retaining selective turn-level weighting. Experiments on agent tasks show that SAGE-OPD consistently improves over baselines, achieving up to a 13.3% relative improvement in ALFWorld unseen success rate over standard OPD. Ablation studies further demonstrate that turn-level intervention, teacher confidence weighting, and loss normalization provide complementary benefits. Our results suggest that effective multi-turn OPD should remain on-policy, but teacher supervision should be selectively allocated to turns where intervention is necessary and reliable.

End-to-end autonomous driving models are typically trained on offline human-demonstration datasets that provide limited state coverage and often no closed-loop feedback, making them prone to compounding errors when deployed in closed-loop and brittle to long-tail agent interactions. To overcome these limitations, we propose an alternative strategy for training end-to-end driving models: large-scale self-play directly from pixels in simulation. While prior self-play approaches have shown promising transfer to real-world driving, they typically assume vectorized Bird's-Eye-View (BEV) observations that are incompatible with end-to-end policies operating directly on sensor observations. To this end, we introduce Gigapixel, a high-throughput batched driving simulator with perspective rendering, enabling scalable self-play directly from pixel observations. Rather than targeting compute-costly photorealistic sensor simulation, Gigapixel renders a simplified bounding-box world that preserves essential scene structure while achieving throughput at 50k agent steps per second. Since direct pixel-space self-play RL is prohibitively sample-inefficient at end-to-end model scale, we propose self-play DAgger training: we train pixel-based policies in self-play via on-policy distillation from a privileged RL teacher. To bridge the sim-to-real gap, we subsequently transfer the self-play trained policies to real-world sensor data through lightweight perception adaptation. Policies trained in Gigapixel and adapted to real-world sensor data achieve competitive performance on the HUGSIM and NAVSIM-v2 benchmarks without human trajectory supervision. Moreover, scaling self-play training yields proportional gains in policy performance, establishing self-play as a practical and scalable strategy for training end-to-end models.

On-policy self-distillation (OPSD) trains a model on its own rollouts and uses a frozen copy to provide dense token-level targets conditioned on a reference target. This works well for LLM reasoning, but a direct extension to multimodal large language models (MLLMs) can create a shortcut: the privileged target may guide tokens mainly based on the text reference target rather than the image. We propose ViGOS, a visually grounded OPSD framework for MLLM post-training. The student first writes a visual description and then reasons toward the final answer. For valid rollouts, an image-only perception teacher supervises the description, while a privileged reasoning teacher supervises the reasoning and final answer on the same student prefix. A reference teacher is used only for invalid rollouts to recover the output format. Across general vision-language, expert reasoning, visual math, spatial grounding, and visual-language-prior benchmarks, ViGOS keeps the main benefits of OPSD and improves image-grounded behavior in shortcut-prone settings.

Unified multimodal models (UMMs) interleave generated ''visual thoughts'' (VTs) with text reasoning to improve spatial tasks. This incurs roughly an order-of-magnitude inference cost from multi-step diffusion. We find this cost yields limited direct benefit. On ThinkMorph, removing or noising VTs barely changes accuracy across nine benchmarks. Once rendered, attention concentrates on the VT regardless of content. Yet a KL diagnostic shows that conditioning on a privileged VT trace shifts the model's completion distribution. This suggests the generation pathway encodes useful reasoning beyond the rendered pixels. Motivated by this gap, we propose Visual On-Policy Self-Distillation(Visual-OPSD). Teacher and student share identical weights but differ in context: the teacher sees privileged VTs while the student sees only the question. Token-level JSD distillation on on-policy student trajectories transfers the teacher's reasoning to a text-only student. Across nine benchmarks, Visual-OPSD improves over its generative teacher by $+3.40$pp with $14.3\times$ speedup (10.0s vs. 142.8s per sample) and outperforms same-scale VLMs by $+63.83$pp on VSP. A Gaussian-noise control ($+0.40$pp vs. $+10.28$pp for real VTs) and $58.4\%$ closure of the KL gap confirm that gains come from the semantic content of the generation pathway.

Knowledge distillation transfers a teacher's competence to a small student but is brittle in the small-student regime: forcing the student to imitate logits from a much larger teacher concentrates it on the teacher's sharpest modes, hurting generalization on benchmark families beyond the training corpus. Reinforcement learning (RL) avoids logit imitation by training on the student's own rollouts. However, on questions where every rollout fails, yielding zero advantage and being silently discarded, injecting a stronger teacher's response into the policy gradient breaks the on-policy assumption and induces drift. We introduce Zone of Proximal Policy Optimization (ZPPO), inspired by Vygotsky's zone of proximal development, which keeps the teacher inside the prompt rather than the policy gradient. On hard questions, ZPPO constructs two reformulated prompts. A Binary Candidate-included Question (BCQ) pairs one correct teacher response with one incorrect student response as anonymized candidates the student must reason between. A Negative Candidate-included Question (NCQ) aggregates the student's wrong rollouts into a single prompt to surface their shared failure modes. A prompt replay buffer recirculates each hard question until it either graduates (the student's mean rollout accuracy on it reaches half or more) or is FIFO-evicted under finite capacity, amplifying BCQ and NCQ inside the student's current zone of proximal development. On the Qwen3.5 family at four student scales (0.8B-9B) with a 27B teacher, post-trained as vision-language models and evaluated on a 31-benchmark suite (16 VLM, 10 LLM, 5 Video), ZPPO outperforms off/on-policy distillation and GRPO, with the largest gains at the smallest scale.

As an increasing majority of global video content is consumed on social platforms for interactive social purposes, video generation models built for social worlds are important but largely overlooked by previous studies. In this work, we define the position of social world models and build a prototype model as the first step towards this goal. While previous world models successfully simulate physical environments or gaming world exploration, they remain fundamentally detached from human-centric social dynamics. To bridge this gap as the first step to social world models, we present MaineCoon, the first real-time audio-visual autoregressive model that has 22B parameters and is capable of real-time streaming generation and sub-second interaction, with a record-breaking frame rate of up to 47.5 FPS, on a single GPU. To the best of our knowledge, MaineCoon is also the first real-time audio-visual generation model specifically optimized for social-interactive applications. To enable efficient and stable training, we introduce several novel techniques into MaineCoon, including self-resampling, cross-modal representation alignment, domain-aware preference optimization, and reinforced online-policy distillation (ROPD). We also design the first agentic streaming inference framework that supports thousand-second-scale or even longer generation while mitigating drift with agentic cache management and prompt planing. These innovations significantly accelerate training while optimizing real-time inference performance. We believe this work not only sets a new state-of-the-art (SOTA) performance benchmark for high-quality, low-latency, and long-horizon audio-visual autoregressive models, but also points out the paradigm shift desired for next-generation AI-native social platforms.

Standard on-policy distillation (OPD) for large language models estimates the reverse-KL objective using student-sampled tokens, yielding an unbiased single-sample Monte Carlo estimator that avoids vocabulary-wide computation. However, we show that this estimator suffers from severe training pathologies in practice: sample inefficiency, unstable generation dynamics, and a substantial performance gap compared to exact full-vocabulary OPD. Reward-level diagnosis traces these pathologies to the log-ratio reward, which is unbounded by construction, producing extremely high-variance gradients concentrated at early positions and persisting throughout training; standard post-hoc scaling fail as they operate only after this distortion occurs. To solve this problem, we propose PowerOPD: a family of natively bounded, sign-consistent rewards from the Box-Cox power transformation, parameterized by alpha > 0, of which the log-ratio is the degenerate alpha -> 0 limit. Across six mathematical reasoning benchmarks and four Qwen3 teacher-student pairs, PowerOPD achieves benchmark-averaged Avg@8/Pass@8 gains of up to +6.37/+5.71 over vanilla OPD, +3.01/+3.54 over post-hoc stabilization, and +2.59/+8.90 over full-vocabulary OPD, while reducing wall-clock time by 59.2% and peak GPU memory by 23.1%. Larger alpha generally improves accuracy, consistently shortens responses, and keeps gradient norms more than 3,000x smaller than vanilla OPD.

Multi-turn agents that plan, invoke tools, and interact with environments offer a promising paradigm for solving complex tasks, yet their capabilities typically rely on very large models whose inference cost is prohibitive in practice.On-Policy Distillation (OPD) is a natural recipe for transferring such capabilities to smaller students, but we find that it suffers a characteristic failure mode in this setting: small student errors compound across turns and push the trajectory out of the teacher's familiar state distribution, so the teacher's supervision becomes least reliable precisely where the student needs it most.We propose Guided On-Policy Distillation (Guided-OPD), a simple yet effective algorithm that mixes teacher- and student-generated turns within each rollout and schedules the teacher's intervention probability along a curriculum that decays to zero.Strong guidance keeps early trajectories close to the teacher distribution and is then gradually withdrawn to recover the purely on-policy regime used at inference.On ALFWorld, ScienceWorld, and WebShop, distilling Qwen3 students from a Qwen3-30B-A3B teacher, Guided-OPD improves Score by 21.1% and Success Rate by 25.5% over vanilla OPD on average, with larger gains on smaller students.

Reinforcement learning from verifiable rewards assigns a single scalar to each rollout, leaving token-level credit assignment underspecified in long reasoning traces. On-policy self-distillation addresses this by letting the same model act as a teacher conditioned on privileged information, producing a dense per-token signal. But the common choice of a ground-truth answer is only an endpoint cue: on terse-answer tasks, the teacher falls silent at the intermediate positions where path-level guidance matters most. We propose Hindsight Self-Distillation (HSD), which conditions the teacher on a successful peer rollout drawn from the current training group. Such a peer is an exact sample from the success-conditioned policy, requiring no additional sampled rollouts. By providing a full successful continuation rather than only the final answer, the resulting credit signal concentrates at the divergence position between a failed rollout and a successful peer. Across Qwen3-8B and Qwen3-32B on math and code benchmarks, HSD obtains the best result against GRPO variants and on-policy distillation baselines, with the largest gains on terse-answer tasks such as AIME.

We introduce Nemotron 3 Ultra, a 550 billion total and 55 billion active parameter Mixture-of-Experts Hybrid Mamba-Attention language model. We pre-trained Nemotron 3 Ultra on 20 trillion text tokens, then extended the context length to 1M tokens, and post-trained using Supervised Fine Tuning (SFT), Reinforcement Learning (RL), and Multi-teacher On-Policy Distillation (MOPD). Nemotron 3 Ultra is our most capable model yet, employing multiple key technologies - LatentMoE, Multi Token Prediction (MTP), NVFP4 pre-training, multi-environment RLVR, MOPD, and reasoning budget control. Nemotron 3 Ultra achieves up to ~6x higher inference throughput as compared to state-of-the-art publicly available LLMs while attaining on-par accuracy. The state-of-the-art accuracy, high inference throughput, and 1M token context length make Nemotron 3 Ultra ideal for long-running autonomous agentic tasks. We open-source the base, post-trained, and quantized checkpoints, along with the training data and recipe on HuggingFace.

On-policy distillation (\textsc{OPD}) has recently become a prominent post-training recipe by combining two desirable ingredients: on-policy student trajectories and dense teacher supervision. However, how this hybrid changes a model's parameters remains unclear. Across several language and vision-language model pairs and \textsc{OPD} use cases, our analysis yields two main findings. On sparsity, \textsc{OPD} updates are small and coordinate-sparse. They are distributed across layers, with the largest relative movement usually appearing in FFN modules. This sparse structure is operationally useful: training only the discovered subnetwork nearly recovers full-training performance. The sparse support does not remove the need for adaptive optimization: SGD, previously reported to be competitive in \textsc{RLVR}, underperforms AdamW in our \textsc{OPD} optimizer ablation, suggesting that dense teacher supervision preserves useful momentum structure and heterogeneous second-moment scales. On geometry, the updates are numerically full-rank but spectrally concentrated; they lie mostly away from the principal singular subspaces of the source weights and fall disproportionately on coordinates where the source weights are close to zero. These findings suggest that dense teacher supervision does not turn \textsc{OPD} into ordinary dense parameter rewriting; instead, \textsc{OPD} retains important geometric signatures of on-policy post-training.

Recent work has shown that on-policy distillation can internalize privileged context, such as system prompts or task hints, into a student model so that the context is no longer needed at inference time. Although this approach successfully improves the student's no-context performance, we identify an interesting and previously unstudied phenomenon: in many settings, reintroducing the original privileged context to the distilled student actually degrades its performance, even on instances it already solves correctly without context. We term this context-induced degradation and argue that robust internalization demands not only matching the teacher's context-conditioned behavior, but also remaining stable when the context is reintroduced, a property we call context removability. Motivated by this observation, we propose a lightweight consistency regularizer that first anchors the student's no-context output via stop-gradient, then penalizes the context-conditioned output for deviating from it via forward KL divergence. This simple addition requires only one extra forward pass per training step, yet it effectively mitigates context-induced degradation and, in many cases, even improves no-context performance. Across 12 configurations spanning diverse domains and model families, our method improves context-conditioned accuracy in the majority of settings, reduces context-induced harm in 11 out of 12 settings, and effectively eliminates response-length inflation. A mechanistic case study further confirms that context removability is achieved at the representation level, with hidden states remaining nearly identical regardless of whether the context is present.

On-policy distillation (OPD) has demonstrated strong empirical gains in enhancing complex reasoning in LLMs by aligning a student model with a teacher's predictive distribution over the student's own trajectories. An emerging variant, Privileged OPD, further strengthens this paradigm by employing a self-teacher model augmented with privileged information, such as oracle traces, to mitigate teacher-student capacity gaps while providing dense, answer-directed supervision. However, current methods treat privileged information as a monolithic imitation target, failing to disentangle locally reachable reasoning steps from future-conditioned oracle signals. Consequently, the student is encouraged to match a hindsight-biased distribution that often falls outside its local predictive support. This reachability mismatch incentivizes the student model to skip valid intermediate reasoning in favor of locally unsupported shortcuts. To resolve this, we introduce Anchored Residual On-Policy Distillation (AR-OPD), a dual-view framework that disentangles privileged supervision. Rather than enforcing strict full-view imitation, AR-OPD establishes a locally compatible anchor using a partially privileged teacher, isolating and injecting oracle foresight as a controlled residual to provide destination-directed guidance. Across diverse reasoning tasks, AR-OPD outperforms full privileged OPD by 2.3 points and SFT by 7.9 points. Crucially, this anchored residual mechanism reduces hindsight leakage by 21.7% and mitigates late-stage drift, yielding up to a 7.2-point advantage on challenging long-horizon trajectories exceeding 768 tokens.

On-policy distillation (OPD) provides dense token-level supervision by asking a teacher to score student-generated rollouts. However, when the student drifts into an unrecoverable prefix, the teacher may locally agree with the degraded state, producing low reverse KL but little corrective training signal. We identify this persistent regime as a low-KL agreement trap. Further analyses show that tokens during and after such traps produce less useful supervision signals. We propose KAT (KL Agreement Trap Termination), an online OPD termination rule that detects persistent low-KL agreement with a dynamic training-adaptive threshold. By filtering weak supervision from degenerate agreement, KAT improves avg@k accuracy by 2.66% and pass@k by 3.43% across four mathematical benchmarks, while reducing average rollout length by 59.73%.

On-Policy Distillation (OPD) has become a core technique in the post-training of Large Language Models (LLMs) for transferring knowledge from domain experts to student models. However, existing OPD distillation methods require teacher and student models to share the same tokenizer, restricting the applicability of OPD within the model series. Current mainstream practice typically employs Supervised Fine-Tuning (SFT) on teacher-generated responses for cross-tokenizer distillation, which fails to capture the rich knowledge embedded in the teacher's probability distribution. In this work, we enable the standard on-policy distillation method to operate across model families, ensuring that high-fidelity token-level signals can propagate across different tokenizers with a precise token-mapping algorithm. Extensive experiments show that cross-tokenizer OPD is significantly more compute-efficient than baselines on various benchmarks. Our results unlock a broader range of teacher-student pairs for OPD, opening up new avenues for adapting and enhancing interactions between LLMs.

On-policy distillation (OPD) trains a student on its own trajectories with dense per-token supervision from a stronger teacher, and often outperforms off-policy distillation and standard reinforcement learning. However, we find that its effectiveness implicitly relies on two assumptions that frequently break in practice: trajectory-level alignment between the student and the teacher, and uniform token-level reliability of the teacher's preferences. We therefore propose Sign-Gated On-Policy Distillation (SG-OPD), which uses a binary verifier as a trust signal for the teacher at two complementary granularities: phased teacher sampling mixes in verifier-endorsed teacher rollouts at cold-start, and a sign-consistency gate extrapolates the distillation update on tokens where the teacher agrees with the verifier-correct direction and interpolates it where it disagrees. Experiments on competition-level mathematical reasoning benchmarks show that SG-OPD consistently outperforms standard OPD, with average gains of 1.98 and 7.50 at the per-sample and per-question levels, respectively.

On-policy distillation (OPD) has recently emerged as an important post-training paradigm. By using a stronger teacher model to provide dense, fine-grained supervision for sampled trajectories, OPD offers a clear advantage over reinforcement learning with verifiable rewards (RLVR), which typically depends on sparse binary or outcome-based environmental feedback. However, naive token-level distillation can suffer from gradient instability, due to magnitude misalignment in outlier states. To address this issue, we propose Globally Normalized Distillation Policy Optimization (GNDPO), a practical method that stabilizes optimization by transforming raw KL scores into batch-level relative advantages. This normalization effectively mitigates gradient explosions while retaining the benefits of token-level guidance. Experimental results show that GNDPO substantially improves training robustness and downstream performance across multimodal reasoning tasks. The code is released at https://github.com/OPPO-Mente-Lab/GNDPO.

On-policy distillation (OPD) is increasingly used to improve large language model reasoning, but its training dynamics remain poorly understood. We characterize the trajectory of OPD updates in parameter space and compare it with supervised fine-tuning (SFT) and reinforcement learning with verifiable rewards (RLVR). A suite of parameter-space diagnostics consistently places OPD in a relaxed off-principal regime: compared with SFT, its updates affect fewer weights and avoid principal directions more strongly, while compared with RLVR, they remain less tightly constrained. Beyond this static localization, OPD exhibits subspace locking: its cumulative updates rapidly enter a narrow low-dimensional channel. Constraining training to the update subspace formed early in training preserves OPD performance but substantially degrades SFT, indicating that the locked subspace is functionally sufficient for OPD. Control experiments further show that sparsifying the update tokens and shifting rollout generation off-policy preserve the rank dynamics, whereas mixing the OPD objective with RLVR changes them. Overall, these results suggest that OPD is not merely an intermediate point between SFT and RLVR, but induces its own update geometry in parameter space.

We study the transformation of autoregressive models (ARLMs) into diffusion language models (DLMs). Rather than pretraining from scratch, prior work replaces the causal attention in ARLMs with bidirectional attention and then trains the resulting model using a DLM objective. However, these approaches incur two distribution shifts. First, transitioning from a next-token prediction objective to a DLM objective can discard knowledge acquired by the ARLM during training. Second, standard DLMs suffer from a train-inference mismatch, as the training loss is defined on randomly masked sequences rather than the trajectories encountered at inference produced by confidence-based decoding. To address both challenges, we introduce an On-Policy Diffusion Language Model (OPDLM) in which On-Policy Distillation (OPD) is employed for ARLM-to-DLM transformation. Specifically, OPDLM is trained via self-OPD, where the student, an ARLM with bidirectional attention, generates its own trajectories, and the teacher, the original frozen ARLM, distills its knowledge by providing target logits on these trajectories. By training directly in an on-policy manner, OPDLM eliminates the train-inference mismatch in DLMs, while distillation from the original model enhances knowledge retention from the ARLM. Empirical results demonstrate that OPDLM requires 15x to 7,000x fewer training tokens with strong performance across a wide variety of tasks. OPDLM avoids the prohibitive cost of DLM pretraining and positions DLM transformation as a form of ARLM post-training.

On-policy distillation (OPD) supervises the student exclusively in the output space by matching next-token distributions. This paradigm suffers from two limitations: (i) a high-variance gradient estimator whose signal-to-noise ratio collapses as the student approaches the teacher, and (ii) an LM-head information bottleneck that discards the teacher's intermediate hidden states. We propose On-Policy Representation Distillation (OPRD), the first method to lift on-policy distillation into the hidden-state space. OPRD aligns student and teacher representations across selected layers on the same on-policy rollouts, providing dense, deterministic, per-layer supervision while bypassing the LM head entirely. Theoretically, OPRD provides a deterministic per-sample gradient, removing the token-level estimation variance that plagues OPD, and exposes structural information that any output-space objective necessarily discards. Empirically, OPRD closes the student-teacher gap on competition mathematics benchmarks (AIME 2024, AIME 2025, and AIMO), where every output-space baseline plateaus below the teacher, while training 1.44x faster and using up to 54% less memory. We further extend OPRD to the cross-architecture setting via OPRD-Bridge. By exploiting the observation that heterogeneous models share a low-rank representational structure, we construct a frozen projector pair that aligns representations across arbitrary depth and width mismatches, shifting the alignment from the output space (which depends on a shared vocabulary) to the representation space. We validate OPRD-Bridge on both cross-architecture (Qwen3-4B -> Qwen3-1.7B-Base) and cross-tokenizer (Phi-4-mini-reasoning -> Qwen3-1.7B-Base) settings, demonstrating successful knowledge transfer even when the vocabulary-based alignment channel is unavailable. Code: https://github.com/ShenzhiYang2000/OPRD.

On-policy distillation (OPD) improves reasoning by training a student on trajectories sampled from its own policy under supervision from a teacher. In multimodal reasoning, a common extension is to use a privileged teacher that observes training-time-only signals such as reference answers or rationales. However, such answer-side privilege creates a train-test mismatch: the teacher's supervision may depend on signals unavailable to the student, encouraging shortcut imitation rather than visually grounded reasoning. We propose ViCuR, a visually grounded privileged-teacher distillation framework that replaces answer-side privilege with visual cues (query-related evidence in the input). Because these cues are derived from the same visual input available at inference, their evidence is recoverable by the student. To support this, ViCuR introduces a lightweight cue recovery module that uses dedicated sink-token cross-attention during prefill to aggregate task-relevant visual evidence into an internal representation, without changing the inference interface or requiring auxiliary cue-generation losses. Across seven benchmarks with Qwen3-VL-2B and 8B students, ViCuR consistently improves over answer-based on-policy self-distillation by +1.19 and +1.24 on overall average performance. It also extends naturally to stronger-teacher OPD, surpassing OPD baselines by +0.64 and +1.08, with consistent out-of-domain gains at the 8B scale. These results show that, in multimodal on-policy distillation, the design of teacher privilege is as important as teacher strength.

Experience internalization converts contextual experience from past interactions into reusable parametric capability, offering a promising path toward continual learning in large language models (LLMs). While prior work has predominantly focused on single-iteration transfer, we discover that under multi-iteration experience learning, existing methods suffer from a progressive capability collapse rather than compounding improvement. We systematically examine this failure through three vital dimensions of experience internalization: (1) Experience Granularity: We find that principle-level experience is more durable than instance-level experience, as it effectively abstracts transferable strategies away from trajectory-specific details. (2) Experience Injection Pattern: Our analysis reveals that step-wise injection significantly outperforms global injection by aligning experience with intermediate decision states, a property that is critical for long-horizon tool use. (3) Internalization Regime: We demonstrate that off-policy context-distillation on high-quality teacher trajectories provides a substantially more stable training signal than on-policy context-distillation, which is inherently limited by local corrections on student-induced flawed states. Together, these insights yield a simple yet robust recipe for stable and sustainable experience internalization, providing concrete guidance for engineering self-evolving and continually learning LLMs.

The past few decades have witnessed significant advances in the design of machine learning algorithms, from early studies on task-specific shallow models to more general deep Large Language Models (LLMs). Despite showing promising results in tasks that require instant prediction or in-context learning, existing models lack the ability to continually learn and effectively transfer their temporal in-context knowledge to their long-term parameters. Inspired by human learning process, we introduce a ''Sleep'' paradigm that allows the models to continually learn, distill their short-term fragile memories into stable long-term knowledge with replay, and recursively improve themselves with ''Dreaming'' process. In more detail, sleep consists of two stages: (1) Memory Consolidation: an upward distillation process, called Knowledge Seeding, where the memories of a smaller-self are distilled into a larger network to provide more capacity while preserving the knowledge. As a proof of concept, we present a new Generalized Distillation process for {Knowledge Seeding} (i.e., the combination of on-policy distillation with Reinforcement Learning (RL)-based imitation learning); (2) Dreaming: a self-improvement phase, where the model uses RL to generate a curriculum of synthetic data to rehearse new knowledge and refine existing capabilities without human supervision. Our experiments on long-horizon, continual learning, knowledge incorporation, and few-shot generalization tasks support the importance of the sleep stage.

As language models increasingly consume one another's outputs, covert influence -- a phenomenon where a sender's payload (the behavioral disposition it is conditioned to propagate) transfers to a receiver through carriers undetectable by humans -- becomes a growing risk. We characterize this risk across three interfaces: supervised fine-tuning, on-policy distillation, and in-context learning, and find that they vary in the scale of influence achievable without leaving behind human-visible traces. Using inference-time per-sample attribution scores, we study covert influence across all three interfaces with the ability to select carriers that amplify training-time influence, unlocking payload transfers that prior work could not achieve. We further provide evidence that covert influence with natural-language carriers is a distinct phenomenon from prior studies using number carriers, as the latter is more resistant to human detection and less portable across model families. Together, these results suggest that the risk surface for covert influence is broader than previously recognized, and we study pointwise attribution scoring methods as a tool to investigate and mitigate it.

Self on-policy distillation trains a student policy against a teacher derived from its own parameter history, yet the teacher's update schedule -- which governs the \emph{temporal coupling} between teacher and student -- has not been systematically studied as a stability variable. Through a controlled schedule sweep on Qwen3-8B, we establish that \emph{isolation periods}, defined as complete teacher freezing between updates, are the key structural property enabling stable learning, not teacher age. To characterize these underlying training dynamics, we introduce a diagnostic framework of temporal KL structure, refresh shock, and length-tail risk. This framework further uncovers \emph{state-oblivious collapse}: optimal short-horizon fixed schedules catastrophically fail under long-horizon training because a clock-driven refresh can copy a transiently drifting student into the teacher in a single, irreversible step. This failure mode is invisible under short-horizon evaluation and mechanistically distinct from EMA's chronic contamination. To address this, we propose \emph{Consolidation-Gated Teacher Refresh} (CGTR), which preserves isolation periods while gating each refresh on joint evidence of reward improvement and length-tail safety, ensuring every teacher movement responds to genuine student consolidation rather than a clock signal. With a single shared parameter set and no per-dataset retuning, CGTR achieves \textbf{zero collapse} and the best final score on all four tasks (Chemistry, Biology, Physics, ToolUse), self-regulating its refresh frequency to each task's learning dynamics.

On-Policy distillation (OPD) in large language models is shifting from full-trace KL supervision toward more selective training paradigms. Recent OPD methods increasingly focus on selecting which trajectories to learn from, which tokens are most informative, and which supervision signals are most reliable. Motivated by this trend, we rethink optimization granularity of OPD and propose \fireicon\ FiRe-OPD (Filter, then Reweight), which jointly adjusts supervision signals at both trajectory and token levels. In details, FiRe-OPD first filters trajectories to remove low-quality rollout samples, and then applies soft reweighting within the retained trajectories to emphasize informative tokens. Compared with hard token selection, FiRe-OPD leverages a soft-weighting mechanism to effectively mitigate information loss and enhance optimization stability, thereby achieving finer-grained OPD optimization. We validate the effectiveness of FiRe-OPD across strong-to-weak, single-teacher, and multi-teacher settings, and demonstrate its superiority over recent token-level OPD methods ( (e.g., +6.25 on AIME 2024 in strong-to-weak, +18.81 on Miner in multi-teacher). Our code is available at https://github.com/YuYingLi0/FiRe-OPD.

Aligning Large Language Models (LLMs) with human values often degrades their general capabilities, termed the alignment tax. Existing methods mitigate this by balancing dual objectives, which heavily rely on massive general-purpose data or auxiliary reward models. In this paper, we argue that, because safety features are inherently sparse within the output distribution, alignment requires localized modifications rather than global trade-offs. To this end, we propose SafeSteer, which performs on-policy distillation confined to safety tokens. First, we construct a safety teacher via activation steering. Based on this teacher, we develop a safety token selection algorithm. Consequently, SafeSteer restricts the reverse KL penalty to these tokens during training to preserve general capabilities. Experimental results across diverse models show that our SafeSteer achieves a superior trade-off between safety and general capability compared with existing methods, attaining strong safety performance on seven safety benchmarks with only minimal degradation on five general capability benchmarks. Notably, SafeSteer requires only 100 harmful samples without using any general-purpose data, less than 1% of what previous baselines used, considerably reducing alignment cost. More details are on our project page at https://anjingkun.github.io/SafeSteer.

On-Policy Distillation (OPD) trains a student model on its own generative trajectories under dense token-level feedback from a stronger teacher, mitigating both the off-policy distribution shift of Supervised Fine-Tuning (SFT) and the sparse credit assignment of Reinforcement Learning (RL). However, standard OPD faces two coupled limitations. First, it requires direct access to the teacher's token-level logits, excluding a broad class of capable proprietary models from serving as teachers. Second, the token-level logit signal itself is brittle, depending on a narrow overlap of plausible next tokens between teacher and student, and prone to amplifying degenerate patterns such as repetition loops. In this paper, we introduce OmniOPD, a novel framework that addresses both limitations through a logit-free, chunk-level supervision signal. OmniOPD replaces deterministic logit matching with Monte Carlo rollouts that approximate the teacher's local preferences through a continuous semantic similarity metric over multi-token chunks, and concentrates this supervision via a peak-entropy scheduler that audits the student only at its high-uncertainty reasoning forks. A Dirichlet-Multinomial Bayesian prior and a base-model KL anchor further bound the variance of discrete sampling and prevent policy collapse across unaudited tokens. Across competitive benchmarks, OmniOPD surpasses the standard OPD approach by up to +28.64% on math, confirming that chunk-level semantic verification extracts a more reliable learning signal than token-level logit matching, whose high information density is offset by significant noise and brittleness. Furthermore, when paired with stronger black-box teachers such as Claude-4.5-Haiku and Gemini-2.5-Flash, OmniOPD achieves an additional +9.54% relative on math over its open-weight teacher counterpart, advancing the student past the performance of self-exploratory RL.

On-Policy Distillation (OPD) is a fundamental technique for efficient post-training of large language models (LLMs), with broad applications in agent learning, multi-task enhancement, and model compression. However, OPD training becomes unstable when the teacher and student distributions differ substantially, as teacher supervision on student-generated tokens may yield unreliable policy gradients and even cause optimization failure. This work addresses reliable on-policy token-level supervision through credit assignment strategies, and proposes Trust Region On-Policy Distillation, TrOPD. It features the following characteristics: 1) Trust-Region On-Policy Learning: TrOPD performs OPD only in regions where the teacher provides reliable supervision, mitigating the optimization difficulty of the K1 reverse-KL estimator under distribution mismatch. 2) Outlier Estimation: For outlier regions, we explore gradient clipping, masking, and forward-KL estimation to reduce the adverse effects of unreliable supervision. 3) Off-Policy Guidance: The student continues generation from teacher prefixes and uses forward KL to imitate off-policy guidance, encouraging on-policy exploration toward reliable regions. Experiments show that TrOPD consistently outperforms SoTA OPD baselines, including OPD, EOPD, and REOPOLD, across mathematical reasoning, code generation, and general-domain benchmarks.

On-policy distillation (OPD) is a widely used technique to transfer capabilities from capable teacher language models to the base student models, and can be formulated in a reinforcement learning style objective using student generated rollouts. Yet, despite the divergence reward being dependent on student model likelihood, existing works usually adopt a stop gradient design primarily for stability, which makes the resulting advantage estimation questionable. In this work, we provide a generic optimization framework based on f-divergence between the student and teacher, and mathematically revisit whether such design space is valid. We prove that general stop-gradient operation would lead to biased estimates of the reward objective and corresponding gradient for general divergence functions. We propose OPD+, the corrected version of OPD that demonstrates improved performance over the baseline KL approach and also supports the choice of various f-divergence. We validate our findings on mathematical reasoning and tool-use benchmarks.

While on-policy distillation offers dense supervision for training small reasoning models, its optimization dynamics in the multimodal domain remain under-explored. In this work, we challenge the standard monolithic view of Vision-Language Model (VLM) distillation by mathematically decomposing the loss into two distinct components: the language prior and visual grounding. Our analysis uncovers that gradient vectors for these components are nearly orthogonal, indicating that the objective of aligning with the teacher's language distribution is geometrically independent from the objective of matching its visual perception. Consequently, standard optimization passively follows a suboptimal compromise trajectory that implicitly balances the two objectives. Hypothesizing that visual grounding constitutes the primary bottleneck for vision-language reasoning, we introduce Visual Gradient Steering (VGS), a method that dynamically reorients the update vector to prioritize the visual subspace. Experimental results on multiple distillation settings and complex multimodal benchmarks demonstrate that VGS significantly outperforms the standard monolithic formulation of on-policy distillation, achieving superior grounding with minimal training overhead.

Deep reinforcement learning (RL) offers a promising route to real-time power grid operation, yet large neural policies are costly to evaluate, hard to deploy on constrained hardware, and opaque to operators. We ask whether a Proximal Policy Optimization (PPO) agent for grid topology control can be compressed into compact tree-based surrogates without losing operational performance. A PPO teacher is trained on Grid2Op's standard 14-bus environment with a stability-oriented reward, using stress-focused data collection on critical, high-loading states. The policy is then distilled into a decision tree and a random forest. Across held-out validation episodes, both surrogates exceed the teacher in mean reward and survival length at a fraction of the inference cost. The decision tree shows high exact-action agreement with the PPO argmax and near-complete agreement within its top-ranked actions, while remaining small enough to be inspected directly. Feature-importance analysis reveals a representational shift: the PPO policy relies mainly on line-loading signals, while the distilled tree is driven primarily by bus-topology variables. These results suggest that stress-focused distillation can convert a black-box neural controller into a lightweight, auditable rule-like surrogate suited for real-time deployment, while also surfacing risks tied to deterministic actions and topology-specific generalization.

As large language models become stronger, weak supervisors may fail to provide reliable labels, preferences, or final judgments for complex outputs, limiting both weak-to-strong generalization and scalable oversight. We study a more tractable form of weak supervision: using a weak model as a critic rather than as a labeler or judge. Instead of solving the task or selecting the correct answer, the weak critic only needs to provide a non-misleading revision direction that helps the strong model better use its own knowledge. We call this setting weak-critic strong oversight. We first show that weak critiques can improve frozen strong models at inference time, and that critique quality is key to this improvement. We then propose progressive on-policy critique distillation (OPCD), which filters high-quality critiques and distills critic-guided behavior into the strong model through adaptive self-teacher signals. Experiments on reasoning and alignment benchmarks show that our method improves strong models over training epochs, suggesting an effective path for scalable oversight with weak supervision.

On-Policy Distillation (OPD) improves large language model reasoning by training a student model on trajectories sampled from its own policy under teacher supervision. Although OPD operates on trajectories, its learning signal remains token-level: it identifies deviations through high-loss tokens and repairs them through local reverse-KL correction. We show that this "trajectory-sampled but token-learned" mechanism cannot reliably bridge student trajectories toward teacher trajectories. About 30% of high-loss tokens fall into the low-divergence regime, indicating that many are surface-form mismatches rather than real reasoning forks. Moreover, even truly divergent tokens are difficult to repair with isolated token-level supervision, since reasoning failures often unfold as short-horizon distributional drift. We propose Trajectory-aware OPD (TOPD), which uses near-future trajectory information to identify real divergent states and distribute guidance across multiple future tokens. Experiments show that suppressing non-divergent high-loss tokens improves standard OPD from 47.8% to 48.2% average accuracy, while TOPD further improves performance to 52.2%, with gains on AIME24 from 60.0% to 63.3% and AIME25 from 46.7% to 53.3%.

On-policy distillation (OPD) provides dense teacher feedback along student-generated rollouts rather than fixed teacher traces and has emerged as a promising post-training paradigm. However, standard OPD typically generates full rollouts during training, which is computationally expensive and may expose the student to unreliable teacher feedback at late rollout positions, especially during early training. We identify the rollout horizon as a key bottleneck in OPD that substantially impacts training efficiency. Unlike Reinforcement Learning with Verifiable Rewards (RLVR), OPD does not require a final answer reward to provide learning signals. Therefore, full rollouts may not always be necessary for OPD. Motivated by this insight, we propose two simple horizon-control strategies: Progressive OPD (POPD), which gradually expands the rollout horizon during training, and Truncated OPD (TOPD), which permanently performs distillation on reliable truncated rollouts. Experiments on mathematical reasoning show that POPD improves the training efficiency of OPD by up to 3$\times$, while TOPD matches OPD performance using only 10% of the rollout horizon, leading to substantial wall-clock and memory reductions. These results demonstrate that controlling the rollout horizon offers a simple and practical path to more efficient OPD.

On-policy distillation (OPD) trains a student on prefixes sampled from its own policy while matching a stronger teacher. This addresses the prefix mismatch of offline distillation, but early student rollouts can still be poor, placing teacher supervision on weak or low-quality prefixes. We propose Trust-Region behavior Blending (TRB), a warmup method that replaces the early rollout policy with the closest-to-teacher behavior policy inside a student-centered KL trust region, while keeping the per-prefix reverse-KL OPD loss unchanged. The KL budget is annealed to zero, so training returns to pure student rollouts after warmup. Across two math-reasoning distillation settings, TRB attains the strongest average among the compared methods.

On-policy distillation transfers reasoning capabilities by training a student model on its own generated trajectories using token-level feedback from a teacher. However, we identify a critical bottleneck, \textbf{Supervision Fidelity Decay (SFD)}: as student-generated prefixes lengthen, the teacher's next-token distribution becomes less confident and less discriminative. Consequently, the teacher-dependent corrective signal in reverse-KL distillation weakens, causing student drift to compound across long reasoning chains. To mitigate SFD, we introduce \textbf{Lookahead Group Reward (\ours{})}. Building on the insight that next-step teacher confidence reflects the discriminative strength of future reverse-KL supervision, \ours{} evaluates the student's top-K candidate tokens by the teacher confidence they induce at the subsequent step and assigns a group-normalized reward. To maintain computational efficiency, we further design an entropy-triggered tree-attention mechanism. Across six math and code benchmarks, \ours{} improves mean@8 by \textbf{2.57} points over OPD for a 7B student, with gains increasing in longer-generation and reaching +\textbf{4.92} points on AIME-26 at 39k tokens.

Large language models (LLMs) often solve a task when all instructions are given in a single prompt, but fail when the same information is revealed gradually across turns. When a clean FULL prompt and a RAW-SHARDED conversation contain the same complete user evidence, the model should still arrive at the same answer. We argue that a key reason for this gap is self-anchored drift: responses produced under partial information introduce unsupported assumptions, and those assumptions later distort the final answer. To reduce this effect, we propose Canonical-Context On-Policy Distillation (CCOPD). During training, the same base model is used in two roles: a frozen teacher conditioned on the clean FULL prompt and a trainable student that receives the same evidence incrementally through a multi-turn conversation; CCOPD aligns the student's behavior on its own trajectories with the teacher's canonical full-context behavior. Trained only on math problem conversations, CCOPD yields a 32% average relative improvement in RAW-SHARDED performance over the original base model across math and five zero-shot out-of-domain task families, while largely preserving full-context performance. Further analyses suggest that CCOPD strengthens grounding in user evidence and reduces sensitivity to contamination from earlier assistant turns.

Reinforcement learning (RL) is a natural fit for agentic knowledge base question answering (KBQA), where a model must issue executable actions, observe knowledge-base feedback, and eventually return an answer. However, current RL-based KBQA systems mainly optimize sparse rewards from the final answer, leaving intermediate action errors weakly supervised. This is especially limiting for logical-form annotated KBQA benchmarks: gold logical forms can be converted into executable action sequences, but existing pipelines use them mainly for warm-start data construction rather than for on-policy RL updates. We propose GAPD, a training-time Gold-Action Policy Distillation framework that adds dense token-level guidance to outcome-based RL. To align gold actions with on-policy student rollouts, GAPD uses MID-ANCHOR MATCHING: it treats the intermediate entities reached during student exploration and gold execution as state anchors, and matches student states to gold states through these explored entity sets. The current policy conditioned on this aligned gold action serves as a stop-gradient teacher, whose token distribution is distilled back to the ordinary student policy over generated action-token spans. GAPD consistently surpasses the current state of the art on WebQSP, GrailQA, and GraphQ.

Speculative decoding accelerates large language model inference by pairing a target model with a lightweight draft model whose proposed tokens are verified in parallel. A common way to build draft models, like EAGLE3 or DFlash is supervised fine-tuning (SFT) on target-generated trajectories. However, we observe that SFT quickly plateaus: the draft model's acceptance length on test data stops improving. The reason is an offline-to-inference mismatch: In SFT, the drafter learns from fixed target-generated trajectories, whereas during speculative decoding it is evaluated on blocks proposed under its own policy. This motivates on-policy distillation (OPD), where the target model supervises the drafter on draft-induced states. Yet OPD remains difficult for draft models, as they cannot reliably roll out complete sequences independently, whereas target-assisted generation makes the collected sequences follow the target distribution and thus eliminates the on-policy signal. We therefore propose Draft-OPD, which uses target-assisted rollout for stable continuations and replays drafting from the verification-exposed error positions. This allows the drafter to learn from target feedback on both accepted and rejected proposals, focusing training on the draft-induced errors that limit speculative acceptance. Experiments show that Draft-OPD achieves over $5\times$ lossless acceleration for thinking models across diverse tasks, improving over EAGLE-3 and DFlash by 23% and 13%.

On-policy distillation (OPD) transfers reasoning behavior by training a student on teacher feedback along student-generated trajectories, but standard full-rollout training ties every update to a costly completion and can over-allocate supervision to late positions with low marginal value for the current student. We revisit this assumption through the useful supervision horizon: student-induced rollouts can drift from teacher-preferred continuations, while aligned prefixes may already preserve the long-horizon OPD update direction. We propose ADWIN, an adaptive-window framework for OPD that treats rollout length as an online admissibility decision, training on short teacher-anchored prefixes while using delayed full-rollout probes to audit prefix--full alignment and adapt the next horizon with staleness control. Across math and code reasoning benchmarks in single-task, multi-task, and strong-to-weak settings, ADWIN improves the accuracy--compute trade-off over full-rollout OPD and prefix-based baselines, reducing end-to-end training cost by up to 4.1 times while achieving comparable or better accuracy.

While Knowledge Editing (KE) enables efficient updates, its dominant Static Fact Overwriting paradigm treats LLMs as discrete databases, forcibly injecting isolated facts. Fracturing pre-trained logical topologies, this triggers Epistemic Dissonance -- a pathology where un-evolved legacy priors force the model to explicitly negate the injected update. Idealized interventions reveal that this is an inherent structural flaw rather than mere algorithmic noise, with a zero-distortion proxy yielding a catastrophic 95.6% self-refutation rate. Given the causally driven nature of real-world knowledge, grounding updates in explicit causal narratives effectively collapses this conflict rate to just 6.6%, underscoring the imperative for a paradigm shift toward Causal Editing. To internalize this evolution, we propose CODE (Causal On-policy Distillation for Editing). By coupling causal bootstrapping with asymmetric on-policy distillation, CODE engraves causal transition logic directly into parametric memory. Experiments on LLaMA-3.1 and Qwen-2.5 show CODE drastically suppresses self-refutation to 1.8% while securing robust multi-hop accuracy (up to 83.5%), seamlessly transforming discrete fact injection into coherent knowledge evolution. Code is available at https://github.com/CrashBugger/CODE.

Building competitive automatic speech recognition (ASR) models usually requires large-scale au- dio supervision, which makes reproduction and specialization expensive. We study Ark-ASR, a 0.6B- parameter audio-conditioned language model trained with 100k hours of speech, and examine whether a strong Qwen-ASR teacher can transfer additional recognition capability through on-policy distillation. Across Mandarin and English ASR benchmarks, the proposed training recipe consistently improves over supervised fine-tuning alone and outperforms the same-scale Qwen3-ASR-0.6B baseline on four of five evaluation sets. This is achieved with only 100k hours of speech, compared with the 20M hours of super- vised audio reported for the Qwen3-Omni AuT encoder. The larger Qwen3-ASR-1.7B remains stronger, but the results show that teacher-guided on-policy training can substantially close the gap for compact ASR models under a much smaller audio budget. A support-overlap diagnostic further suggests that the teacher-data stage improves local student-teacher compatibility, matching recent analyses of when on-policy distillation is effective.

Reinforcement learning for multi-turn agents suffers from a credit-assignment mismatch: rewards are sparse and trajectory-level, while success often hinges on a few local decisions. Existing online policy distillation (OPD) provides denser token-level supervision, but typically treats heterogeneous agent trajectories as monolithic strings rather than causal interaction units. We present StepOPSD, a post-rollout preference self-distillation framework that takes the agent step as the unit of credit redistribution. StepOPSD decomposes trajectories into action-centered step segments, rescoring them under hindsight-enriched teacher contexts and converting token-level log-probability gaps into sign-preserving advantage shaping with a normalized per-step credit budget before the GRPO update. Across ALFWorld and Search-QA with Qwen3-1.7B and Qwen2.5-3B-Instruct, StepOPSD attains best or second-best results on subsets most sensitive to local causal errors, including first-place performance on ALFWorld Heat (79.1%), PickTwo (95.0%), Search-QA TriviaQA (61.6%), and tied-best performance on HotpotQA (40.4%). The results further reveal a consistent two-knob law: smaller α_clip acts as a broadly stabilizing local trust region, whereas the optimal global mixing strength λ_mix remains task-dependent. These findings suggest that step-aware distillation is most useful when trajectory-level rewards are weakly aligned with the local action that determines downstream success.

Domain specialization can improve LLM behavior, but often weakens the general capabilities inherited from the original model. Recent Multi-Teacher On-Policy Distillation (MOPD) pipelines recover model capabilities by supervising student-generated trajectories with teacher feedback, but typically assume teacher-aligned prompt coverage, requiring prompts to match the teachers' training distributions. This assumption is difficult to satisfy when the general teacher is an open-source model whose post-training data are unknown. Instead of attempting to reconstruct this hidden distribution, we study general capability recovery with readily available proxy general prompts. We identify two failure modes of vanilla MOPD in this incomplete-coverage situation: recovery-preservation counteraction from mixing conflicting recovery and preservation gradients, and weak-signal flattening from uniformly averaging samples with unequal correction demand. We propose \textbf{Counteraction-Aware Multi-Teacher On-Policy Distillation} (\textbf{CaMOPD}), which addresses these issues with decoupled alternating training and gap-based sample selection. CaMOPD allocates dedicated updates to general recovery, periodically performs domain-preservation updates, and selects samples with larger averaged token-level teacher-student log-probability gaps to concentrate correction signals. Across role-play dialogue and medical reasoning QA scenarios, CaMOPD outperforms all baselines in general capability recovery while maintaining domain-specific behavior. Gradient coherence analyses further support the intended effect of CaMOPD in producing more coherent correction signals.

Learning from demonstrations in embodied control is often cast as behavioral cloning, and recent diffusion or flow-matching policies improve this paradigm by modeling multi-modal expert actions. Yet these methods remain offline supervised learners: the policy is trained only on expert states and receives no corrective signal on the states it actually visits. On-policy distillation (OPD) offers a natural remedy, but standard OPD assumes a strong fixed teacher, which is unavailable in demonstration-only control. We propose \textbf{FA-OPD}, an \emph{adversarial dual on-policy distillation} method in which a Flow Matching (FM) teacher is learned from demonstrations and co-trained with a lightweight MLP student. The teacher provides two complementary signals on student rollouts. The reward channel learns an expert-likeness objective over state-action pairs and drives online exploration through long-horizon policy optimization. The action channel supplies dense local targets at student-visited states, stabilizing exploitation. FA-OPD couples them so that reward distillation enables generalization beyond point-wise demonstrations, while action distillation keeps exploration anchored near expert-like behavior. Across six robot navigation, manipulation, and locomotion benchmarks, FA-OPD beats strong baselines and shows much stronger robustness under noisy or limited demonstrations. Source code: https://github.com/vanzll/FA-OPD.

On-policy distillation has recently emerged as a promising alternative to standard sequence-level imitation, training a student by scoring its own rollouts with a teacher model. However, we observe ``Off-policy Teacher Decay'' problem in this paradigm: for the later tokens, with student's earlier trajectory as context that is off-policy to the teacher, the teacher's ability to produce a corrective score would decay, and may fall back to token-completion behavior learned in the pre-training stage. We empirically verify this problem, and we propose Early Stopping Rollout (ESR) to fix it: a simple yet effective distillation strategy that simply restricts the rollout generation to the first response tokens. We show that ESR both surpasses the full rollout OPD performance across model size, family, tasks and training regime, and exhibit much higher GPU efficiency and training stability, especially under cross model family scenarios. We further investigate the mechanism behind this surprising performance and discovered "Cascading Alignment" and "Sub-mode Commitment" effect of ESR that may explain why it works effectively and even sometimes exceeding the teacher model performance. Besides, we show that this position-based token selection strategy cannot be fully explainable by KL divergence and entropy signals.

On-policy distillation (OPD) trains a student on its own rollouts with token-level teacher supervision. Recent selective OPD methods exploit the non-uniformity of OPD signals by prioritizing high-entropy or high-disagreement tokens. We revisit this principle and ask: which token-level teacher signals are actually learnable? Using a fixed-context diagnostic that measures same-context teacher-student KL reduction, we show that raw KL disagreement is a coarse proxy for learning value. It conflates learnable disagreement, where the teacher assigns corrective mass to the student's top-K candidates, with incompatible disagreement, where the teacher places mass mostly off the student's current support. We formalize this local compatibility as token teachability and show that it better predicts fixed-context improvement than raw KL alone. Motivated by this finding, we propose Teachability-Aware OPD (TA-OPD), a lightweight token-position selection method that applies OPD loss to high-teachability positions without reward models or verifiers. Across Qwen2.5 and Qwen 3 teacher-student settings, TA-OPD often surpasses full-token OPD with only 5% retained tokens and improves over entropy- and divergence-based baselines. Our results reframe selective OPD as selecting learnable teacher signals rather than merely salient tokens.

Reinforcement learning for large language models faces a fundamental trade-off between sample efficiency and asymptotic performance: strictly on-policy methods discard trajectories after a single update, while off-policy reuse introduces distribution mismatch that existing trust-region techniques mitigate primarily by enforcing conservative optimization, often leaving rich training signals underutilized. To investigate this, we perform extensive off-policy updates on fixed data. Our experiments reveal that aggressive multi-step optimization brings rapid initial gains, but excessive updates cause trajectory probabilities to deviate and entropy to collapse, with performance plateauing early. Tightening KL constraints merely lowers the ceiling without resolving the degradation. This motivates Extreme Region Policy Distillation (ERPD), a two-stage framework that decouples sample efficiency from KL efficiency. The first stage performs weakly constrained off-policy optimization on fixed data to maximally extract training signals. The resulting policy provides token-level supervision. In the second stage, we distill these signals into the base policy under trust-region constraints, filtering harmful drift while preserving useful signals. The distilled policy achieves comparable or better performance with substantially smaller KL divergence, indicating that much of the first-stage divergence was spent on unnecessary drift rather than genuine improvement. Crucially, ERPD accommodates both strong and weak teachers: when aggressive optimization yields no stronger policy, even degenerate teachers provide effective supervision via alternative signal construction strategies. We validate ERPD on mathematical reasoning, showing gains for strong base models where on-policy training plateaus, and reliable improvements with weak teachers.

Customized image editing aims to equip pre-trained diffusion models with specific visual effects using limited paired data, typically via Low-Rank Adaptation (LoRA). As the number of desired effects grows, storing and dynamically loading numerous these effect LoRAs significantly increases deployment overhead. Furthermore, current pipelines typically cascade these effect LoRAs with acceleration modules for fast generation, which triggers severe parameter interference and results in concept bleeding and style degradation. We propose CollectionLoRA, a multi-teacher on-policy distillation framework capable of distilling the concepts of up to 50 different effect LoRAs along with few-step generation capabilities into a single LoRA. This fundamentally resolves the feature interference issue and significantly reduces deployment costs. Specifically, the method introduces (i) a Probabilistic Dual-Stream Routing mechanism that enables the model to randomly switch between data sources during training, effectively enhancing its generalization in unseen scenarios; (ii) an Asymmetric Orthogonal Prompting strategy to achieve concept isolation within the prompt space; (iii) a Coarse-to-Fine Distillation Objective to mitigate the distribution gap between the teacher and student models. Extensive evaluations show that CollectionLoRA distills all customized effects and few-step generation into a single LoRA, reducing deployment overhead while achieving concept fidelity comparable to or better than independently trained teacher models. Code: https://github.com/Qwen-Applications/CollectionLoRA

Diffusion models excel at generating diverse and multimodal trajectories for robotic planning, yet their iterative denoising process introduces latency that is incompatible with high-frequency closed-loop control. To address this problem, we propose Dynamic Neural Koopman Distillation, a framework that distills multistep diffusion inference into a single forward pass while retaining the multimodal expressivity of the teacher model. Specifically, we introduce a Factorized Dynamic Koopman layer that models the denoising process through a factorized latent transition with state-dependent modal gains. We evaluate the proposed method on standard D4RL MuJoCo locomotion benchmarks and a physical Kinova manipulator, comparing against one-step baselines. The results show that our method significantly outperforms existing one-step distillation approaches on the reported locomotion tasks, and reduces the inference latency to the millisecond regime compared with the teacher policy. Hardware experiments further demonstrate that our method enables smooth and fast closed-loop execution while maintaining task success and comparable accuracy. A project page is available at https://fdkoopman.github.io/.

On-Policy Distillation (OPD) has gained wide attraction as an LLM post-training paradigm due to its effectiveness in improving capabilities without introducing model distribution drift, and consequently, regression in general tasks. On-Policy Self-Distillation (OPSD) is an efficient use-case of OPD, which is appealing as it requires only a single model as a student and teacher, and it also has the benefit of providing privileged context that is a absent at inference time (e.g. a persona, a private fact, or a worked solution) to the teacher during the training process. The challenge in this approach is that the privileged information can change model behavior more than intended: it can modify reasoning, degrade general capabilities, and affect performance indicators like response length, style, or local token preferences. Consequently, OPSD may train the student on side effects rather than a desired, transferable behavior. In this paper, we study this problem in a rare-token/identity setting and propose EviDence GuidEd On-Policy Distillation (EDGE-OPD), a modification of OPSD with two distinct characteristics: a) it uses guided rollouts to inject privileged-context behavior to the student at sampling time, so that the rare target behavior is actually present in the on-policy data, and b) it applies an evidence mask: the student is updated only at token positions where the privileged context supports the sampled token, rather than on every token in the rollout. We empirically show that OPSD (and its variant RLSD, with and without a verifier) completely fail to learn a target identity, while the integration of guided rollouts allows them to succeed. Additionally, mask-region ablations show that the persona signal is localized to the positive-evidence tail, allows us to draw valuable insights about efficient knowledge transfer and preservation of general purpose capabilities.

Large language model post-training methods such as supervised fine-tuning (SFT), reinforcement learning (RL), and distillation are often analyzed through their loss functions: maximum likelihood, policy gradients, forward KL, reverse KL, or related objective-level variants. We study a complementary factor: the state distribution on which supervision is applied. For an autoregressive policy, a state is a prompt plus generated prefix. SFT trains on fixed dataset states, while RL and on-policy distillation (OPD) train on states induced by the current learner. We formalize post-training as state-distribution shaping and run a controlled smallscale study using Qwen3-0.6B-Base on GSM8K, with TruthfulQA and MMLU as retention evaluations. Our results show three phenomena. First, a mild SFT run improves GSM8K with little forgetting, while a stress SFT run causes substantial retention loss. Second, OPD from a degraded SFT teacher surpasses that teacher on GSM8K, TruthfulQA, and MMLU, despite using the teacher as its only supervision source. Third, a lightweight on-policy RL run improves GSM8K while preserving retention. These results support a state-centric view of post-training: the source and locality of training states can be as important as the form of the supervision signal.

Reinforcement learning with verifiable rewards has become the standard recipe for improving LLM reasoning, but the dominant algorithm GRPO assigns a single trajectory-level advantage to every token, diluting the signal at pivotal reasoning steps and injecting noise at uninformative ones. Critic-free alternatives derived from on-policy distillation supply per-token signals through oracle-conditioned likelihood ratios, yet apply each signal in isolation from the trajectory-level evidence accumulated up to that position. We propose Oracle-Prompted Policy Optimization (OPPO), which rests on a single observation: the oracle signal used by prior distillation-style methods for local discrimination is also the natural Bayesian update of the model's belief about eventual success. Accumulating the signal along a trajectory yields, in closed form and at the cost of one extra forward pass, a running estimate of the success probability at every position, together with a token-level advantage that requires no learned value network and no additional rollouts. A first-order analysis factorizes the advantage into the per-token discrimination signal used by distillation methods modulated by a state weight that concentrates credit on genuinely pivotal tokens, with a directional variance-reduction guarantee. The framework admits two estimators differing only in which model scores the evidence: a \textit{self-oracle} that reuses the student and recovers the on-policy distillation reward as a strict special case, and a \textit{teacher-oracle} that delegates scoring to a stronger frozen model. On two base LLMs across seven mathematics, science, and code reasoning benchmarks, OPPO improves over GRPO, DAPO, and SDPO by up to $+6.0$ points on AMC'23 and $+5.2$ points on AIME'24, with gains that widen monotonically with response length.

On-policy self-distillation (OPSD) trains a student on its own rollouts using a privileged teacher, but its standard objective weights all generated tokens equally, implicitly treating the privileged teacher target as equally reliable at every student-visited prefix. Existing entropy-based OPD methods relax this uniformity by modulating token-level supervision with teacher entropy, but high teacher entropy in reasoning has an ambiguous reliability meaning: it can reflect either non-viable uncertainty or benign solution diversity. To identify this phenomenon, we introduce a branch-viability diagnostic. Specifically, we record next-token alternatives from the privileged-answer teacher prompt, force each alternative after the student prompt plus its on-policy spine prefix, and test whether the resulting student-template continuation recovers the correct answer. On Qwen3-4B, we find that an oriented within-sequence position score is the strongest tested predictor of teacher-token reliability, reaching an area-under-ROC-curve (AUROC) of 0.83; local uncertainty scores are at most 0.57. Motivated by this trajectory-level structure, we propose Position-Weighted On-Policy Self-Distillation (PW-OPSD), which applies an increasing position weight while keeping the same student rollout, privileged teacher pass, and clipped forward-KL target as OPSD. In our comprehensive evaluations with different random seeds, the diagnostic-derived PW-OPSD improves AIME 2024 and AIME 2025 Avg@12 by +1.0 and +1.1 points, and a generalization evaluation on two larger-scale models from different families, DeepSeek-R1-Distill-Llama-8B and Olmo-3-7B-Think, also demonstrates consistent aggregate Avg@12 improvements. These results show that teacher-token reliability in reasoning distillation is trajectory-structured and can be utilized without additional teacher computation.

Multimodal Large Language Models (MLLMs) still struggle with fine-grained visual understanding, where answers often depend on small but decisive evidence in the full image. We observe a regional-to-global perception gap: the same MLLM answers fine-grained questions more accurately when conditioned on evidence-centered crops than on the corresponding full images, suggesting that many failures stem from difficulty to focus on relevant evidence rather than insufficient local recognition ability. Motivated by this observation, we propose Vision-OPD (Vision On-Policy Distillation), a regional-to-global self-distillation framework that transfers the model's own privileged regional perception to its full-image policy. Vision-OPD instantiates two conditional policies from the same MLLM: a crop-conditioned teacher and a full-image-conditioned student. The student generates on-policy rollouts, and Vision-OPD minimizes token-level divergence between the teacher and student next-token distributions along these rollouts. This enables the model to internalize the benefit of visual zooming without external teacher models, ground-truth labels, reward verifiers, or inference-time tool use. Experiments on multiple fine-grained visual understanding benchmarks show that Vision-OPD models achieve competitive or superior performance against much larger open-source, closed-source, and "Thinking-with-Images" agentic models. The code is available at https://github.com/VisionOPD/Vision-OPD

Scaling on-policy distillation (OPD) for large language models (LLMs) confronts a fundamental tension: asynchronous execution is necessary for system efficiency, but structurally deviates from the ideal on-policy objective. To address this challenge, we theoretically decompose the objective discrepancy into rollout drift and supervision drift, capturing staleness in student rollout and teacher context, respectively. Building on this, we introduce a sample-level freshness score that quantifies the reliability of a buffered sample with respect to the on-policy objective. Guided by this signal, we further propose f-OPD, a novel framework that adaptively regulates stale-sample influence and constrains policy drift accumulated under asynchronous training. Across reasoning, tool-use, and coding-agent tasks of increasing interaction horizon, f-OPD consistently achieves task performance comparable to synchronous optimization while largely retaining the throughput advantages of asynchronous execution. Our results establish the first recipe for achieving a performance-efficiency trade-off in OPD, paving the way for long-horizon agentic post-training at scale.