Guardado en:
| Autores principales: | , , , , , , , , , , , , , , , , , , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2026
|
| Materias: | |
| Acceso en línea: | https://arxiv.org/abs/2605.02913 |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866914528424361984 |
|---|---|
| author | Surana, Rohan Mundada, Gagan Jiang, Xunyi Wang, Chuhan Tang, Zhenwei Jiao, Difan Huang, Zihan Xiong, Yuxin Wu, Junda Yu, Sheldon Li, Xintong Jain, Raghav Kuang, Nikki Zhou, Sizhe Jin, Bowen Chu, Zhendong Yu, Tong Rossi, Ryan Huang, Kuan-Hao Shang, Jingbo Han, Jiawei McAuley, Julian |
| author_facet | Surana, Rohan Mundada, Gagan Jiang, Xunyi Wang, Chuhan Tang, Zhenwei Jiao, Difan Huang, Zihan Xiong, Yuxin Wu, Junda Yu, Sheldon Li, Xintong Jain, Raghav Kuang, Nikki Zhou, Sizhe Jin, Bowen Chu, Zhendong Yu, Tong Rossi, Ryan Huang, Kuan-Hao Shang, Jingbo Han, Jiawei McAuley, Julian |
| contents | 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. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2605_02913 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Generate, Filter, Control, Replay: A Comprehensive Survey of Rollout Strategies for LLM Reinforcement Learning Surana, Rohan Mundada, Gagan Jiang, Xunyi Wang, Chuhan Tang, Zhenwei Jiao, Difan Huang, Zihan Xiong, Yuxin Wu, Junda Yu, Sheldon Li, Xintong Jain, Raghav Kuang, Nikki Zhou, Sizhe Jin, Bowen Chu, Zhendong Yu, Tong Rossi, Ryan Huang, Kuan-Hao Shang, Jingbo Han, Jiawei McAuley, Julian Machine Learning 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. |
| title | Generate, Filter, Control, Replay: A Comprehensive Survey of Rollout Strategies for LLM Reinforcement Learning |
| topic | Machine Learning |
| url | https://arxiv.org/abs/2605.02913 |