arXiv:2605.02913cs.LG2026-05综述被引 6

系统梳理大模型强化学习中的生成、过滤、控制与重放策略,提升推理训练效率与可靠性。

Generate, Filter, Control, Replay: A Comprehensive Survey of Rollout Strategies for LLM Reinforcement Learning

论文配图:Generate, Filter, Control, Replay: A Comprehensive Survey of Rollout Strategies for LLM Reinforcement Learning
图 1 · 摘自论文原文
  • 将滚动策略拆解为生成-过滤-控制-重放四阶段,统一建模流程
  • 提出可靠性、覆盖性、成本敏感性三维度评估指标,揭示策略权衡
  • 适用于需要高效推理训练的智能体开发与可复现性研究

强化学习已成为提升大语言模型推理能力的关键后训练工具。在这些系统中,滚动(rollout)——从提示到终止所采样的轨迹,包含中间推理步骤及可选的工具或环境交互——决定了优化器学习的数据,但其设计常被忽视。本综述提供了一种面向优化器无关的、基于强化学习的推理型大模型后训练中的滚动策略视角。我们以统一符号形式化滚动流程,并引入生成-过滤-控制-重放(GFCR)生命周期分类法,将滚动流程分解为四个模块化阶段:生成提出候选轨迹与拓扑;过滤通过验证器、裁判、评论者构建中间信号;控制在计算预算下分配算力并做出继续/分支/停止决策;重放保留并跨滚动复用成果(如自演化课程,自主生成新训练任务)。我们辅以可靠性、覆盖性、成本敏感性三类准则,刻画滚动策略间的权衡。在此框架下,我们整合了涵盖可验证奖励强化学习、过程监督、裁判门控、引导式与树/段滚动、自适应算力分配、提前退出与部分滚动、吞吐量优化以及重放/重构用于自我提升的方法。通过数学、代码/SQL、多模态推理、工具使用代理和代理技能基准等案例研究,评估技能归纳、复用与跨任务迁移。最后,我们提供诊断索引,将常见滚动路径问题映射至GFCR模块及其缓解手段,并指出构建可复现、算力高效、可信滚动流程的开放挑战。

原文摘要 · Abstract (English)

Reinforcement learning (RL) has become a central post-training tool for improving the reasoning abilities of large language models (LLMs). In these systems, the rollout, the trajectory sampled from a prompt to termination, including intermediate reasoning steps and optional tool or environment interactions, determines the data the optimizer learns from, yet rollout design is often underreported. This survey provides an optimizer-agnostic view of rollout strategies for RL-based post-training of reasoning LLMs. We formalize rollout pipelines with unified notation and introduce Generate-Filter-Control-Replay (GFCR), a lifecycle taxonomy that decomposes rollout pipelines into four modular stages: Generate proposes candidate trajectories and topologies; Filter constructs intermediate signals via verifiers, judges, critics; Control allocates compute and makes continuation/branching/stopping decisions under budgets; and Replay retains and reuses artifacts across rollouts without weight updates, including self-evolving curricula that autonomously generate new training tasks. We complement GFCR with a criterion taxonomy of reliability, coverage, and cost sensitivity that characterizes rollout trade-offs. Using this framework, we synthesize methods spanning RL with verifiable rewards, process supervision, judge-based gating, guided and tree/segment rollouts, adaptive compute allocation, early-exit and partial rollouts, throughput optimization, and replay/recomposition for self-improvement. We ground the framework with case studies in math, code/SQL, multimodal reasoning, tool-using agents, and agentic skill benchmarks that evaluate skill induction, reuse, and cross-task transfer. Finally, we provide a diagnostic index that maps common rollout pathologies to GFCR modules and mitigation levers, alongside open challenges for building reproducible, compute-efficient, and trustworthy rollout pipelines.

强化学习大模型推理滚动策略智能体

Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品;中文卡片由大模型生成,请以原文为准。