Large Language Model Post-Training: A Unified View of Off-Policy and On-Policy Learning

• URL: http://arxiv.org/abs/2604.07941v2
• Date: Thu, 16 Apr 2026 04:43:04 GMT
• Title: Large Language Model Post-Training: A Unified View of Off-Policy and On-Policy Learning
• Abstract summary: Post-training has become central to turning pretrained large language models into aligned, capable, and deployable systems.<n>Recent progress spans supervised fine-tuning (SFT), preference optimization, reinforcement learning (RL), process supervision, verifier-guided methods, distillation, and multi-stage pipelines.<n>This survey argues that LLM post-training is best understood as structured intervention on model behavior.
• Score: 37.29007534251622
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Post-training has become central to turning pretrained large language models (LLMs) into aligned, capable, and deployable systems. Recent progress spans supervised fine-tuning (SFT), preference optimization, reinforcement learning (RL), process supervision, verifier-guided methods, distillation, and multi-stage pipelines. Yet these methods are often discussed in fragmented ways, organized by labels or objectives rather than by the behavioral bottlenecks they address. This survey argues that LLM post-training is best understood as structured intervention on model behavior. We organize the field first by trajectory provenance, which defines two primary regimes: off-policy learning on externally supplied trajectories and on-policy learning on learner-generated rollouts. We then interpret methods through two recurring roles -- effective support expansion, which makes useful behaviors more reachable, and policy reshaping, which improves behavior within already reachable regions -- together with a complementary systems-level role, behavioral consolidation, which preserves, transfers, and amortizes useful behavior across stages and model transitions. Under this view, SFT may serve either support expansion or policy reshaping; preference optimization is usually off-policy reshaping, though online variants move closer to learner-generated states. On-policy RL often improves behavior on learner-generated states, but stronger guidance can also make hard-to-reach reasoning paths reachable. Distillation is often better understood as consolidation rather than only compression, and hybrid pipelines emerge as coordinated multi-stage compositions. Overall, the framework helps diagnose post-training bottlenecks and reason about stage composition, suggesting that progress increasingly depends on coordinated systems design rather than any single dominant objective.

Related papers

• Optimizing What Policies Learn From: Recoverability-aware Rollout Intervention Learning [21.808015281498076]
  Critic-free group-based reinforcement learning has become a scalable approach for post-training large language models.<n>Recent work has started to treat rollout generation as an adaptive decision, but two important limitations remain.<n>We propose Recoverability-Aware Learning (RAIL), a training-time framework that learns how to generate rollouts based on the improvement produced by each intervention.
  arXiv  Detail & Related papers  (2026-08-05T17:22:02Z)
• Self-Improving Large Language Models via Progressive Experience Evolution [52.03479747358687]
  We propose textbfSPEE (textbfSelf-textbfProgressive textbfExperience textbfEvolution), a unified post-training framework that sequentially performs explicit experience evolution followed by implicit policy optimization.<n> Experiments on five mathematical reasoning benchmarks demonstrate that SPEE consistently outperforms both test-time and training-time self-evolution baselines across three model scales.
  arXiv  Detail & Related papers  (2026-08-03T12:27:32Z)
• On-Policy and Off-Policy Learning for Large Action Spaces [2.995087247817663]
  This thesis studies policy learning in interactive systems where an agent observes a context, selects an action from a very large set, and receives partial feedback.<n>The main framework is contextual bandits, with two paradigms: on-policy learning, where the agent interacts sequentially with the environment and minimizes regret, and off-policy learning, where it learns from logged data collected by a logging policy.
  arXiv  Detail & Related papers  (2026-07-30T15:56:11Z)
• SEED: Self-Evolving On-Policy Distillation for Agentic Reinforcement Learning [38.54413500261524]
  Outcome-based reinforcement learning (RL) provides a practical optimization paradigm, but its sparse trajectory-level rewards offer limited guidance on intermediate decisions.<n>We propose SEED (SElf-Evolving On-Policy Distillation), a self-evolving framework that converts completed on-policy trajectories into training-time hindsight skills.
  arXiv  Detail & Related papers  (2026-07-16T09:57:18Z)
• Beyond One-Size-Fits-All: Diagnosis-Driven Online Reinforcement Learning with Offline Priors [59.11261091345906]
  We argue that the field should shift to diagnosis-driven tension management, in which deployment-specific evidence guides how the learner relates to its priors throughout training, enabling both flexible and adaptive deployment.<n>We support this position with a framework characterizing how priors reshape online optimization through three functional roles, controlled experiments demonstrating help-or-hurt reversals, cross-domain evidence from foundation model post-training to embodied intelligence, and engagement with five substantive counterarguments.
  arXiv  Detail & Related papers  (2026-06-24T08:05:28Z)
• LAFP: Preserving Latent Action Structure in Latent Policy Learning via Flow Matching [15.922360971421654]
  We propose Latent Action Flow Policy (LAFP) for scalable latent policy learning.<n>LAFP consistently outperforms prior methods on downstream imitation learning tasks.<n>LAFP incurs less than 1x additional inference overhead.
  arXiv  Detail & Related papers  (2026-06-09T07:49:49Z)
• A Systematic Post-Train Framework for Video Generation [76.26555417456773]
  Large-scale video diffusion models have demonstrated impressive capabilities in generating high-resolution and semantically rich content.<n>A significant gap remains between their pretraining performance and real-world deployment requirements due to critical issues such as prompt sensitivity, temporal inconsistency, and prohibitive inference costs.<n>We propose a comprehensive post-training framework that systematically aligns pretrained models with user intentions through four synergistic stages.
  arXiv  Detail & Related papers  (2026-04-28T09:34:51Z)
• Decouple to Generalize: Context-First Self-Evolving Learning for Data-Scarce Vision-Language Reasoning [41.523848964102]
  Recent vision-language models (VLMs) achieve remarkable reasoning through reinforcement learning (RL)<n>RL provides a feasible solution for realizing continuous self-evolving large vision-language models (LVLMs) in the era of experience.<n>Existing strategies such as synthetic data and self-rewarding mechanisms suffer from limited distributions and alignment difficulties.<n>We propose DoGe, a dual-decoupling framework that guides models to first learn from context rather than problem solving.
  arXiv  Detail & Related papers  (2025-12-07T13:17:31Z)
• Feature-Based vs. GAN-Based Learning from Demonstrations: When and Why [50.191655141020505]
  This survey provides a comparative analysis of feature-based and GAN-based approaches to learning from demonstrations.<n>We argue that the dichotomy between feature-based and GAN-based methods is increasingly nuanced.
  arXiv  Detail & Related papers  (2025-07-08T11:45:51Z)
• Diffusion Guidance Is a Controllable Policy Improvement Operator [98.11511661904618]
  CFGRL is trained with the simplicity of supervised learning, yet can further improve on the policies in the data.<n>On offline RL tasks, we observe a reliable trend -- increased guidance weighting leads to increased performance.
  arXiv  Detail & Related papers  (2025-05-29T14:06:50Z)
• LLM-Guided Reinforcement Learning: Addressing Training Bottlenecks through Policy Modulation [7.054214377609925]
  Reinforcement learning (RL) has achieved notable success in various domains.<n>Training effective policies for complex tasks remains challenging.<n>Existing approaches to mitigate training bottlenecks fall into two categories.
  arXiv  Detail & Related papers  (2025-05-27T03:40:02Z)
• DUMP: Automated Distribution-Level Curriculum Learning for RL-based LLM Post-training [25.91869315787235]
  We present a principled curriculum learning framework grounded in the notion of distribution-level learnability.<n>Our framework prioritizes distributions with either high average advantage (exploitation) or low sample count (exploration)<n>Our experiments show that our framework significantly improves convergence speed and final performance.
  arXiv  Detail & Related papers  (2025-04-13T20:10:27Z)
• Without Paired Labeled Data: End-to-End Self-Supervised Learning for Drone-view Geo-Localization [20.603433987118837]
  Drone-view Geo-Localization (DVGL) aims to achieve accurate localization of drones by retrieving the most relevant GPS-tagged satellite images.<n>Existing methods heavily rely on strictly pre-paired drone-satellite images for supervised learning.<n>We propose a novel end-to-end self-supervised learning method with a shallow backbone network.
  arXiv  Detail & Related papers  (2025-02-17T02:53:08Z)
• Continual Task Learning through Adaptive Policy Self-Composition [54.95680427960524]
  CompoFormer is a structure-based continual transformer model that adaptively composes previous policies via a meta-policy network. Our experiments reveal that CompoFormer outperforms conventional continual learning (CL) methods, particularly in longer task sequences.
  arXiv  Detail & Related papers  (2024-11-18T08:20:21Z)
• The Gradient Convergence Bound of Federated Multi-Agent Reinforcement Learning with Efficient Communication [20.891460617583302]
  The paper considers independent reinforcement learning (IRL) for collaborative decision-making in the paradigm of federated learning (FL) FL generates excessive communication overheads between agents and a remote central server. This paper proposes two advanced optimization schemes to improve the system's utility value.
  arXiv  Detail & Related papers  (2021-03-24T07:21:43Z)
• Hierarchical Variational Imitation Learning of Control Programs [131.7671843857375]
  We propose a variational inference method for imitation learning of a control policy represented by parametrized hierarchical procedures (PHP) Our method discovers the hierarchical structure in a dataset of observation-action traces of teacher demonstrations, by learning an approximate posterior distribution over the latent sequence of procedure calls and terminations. We demonstrate a novel benefit of variational inference in the context of hierarchical imitation learning: in decomposing the policy into simpler procedures, inference can leverage acausal information that is unused by other methods.
  arXiv  Detail & Related papers  (2019-12-29T08:57:02Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.