Single-Rollout Asynchronous Optimization for Agentic Reinforcement Learning

  • 类型:arxiv
  • 标识:2607.07508
  • 链接:https://arxiv.org/abs/2607.07508
  • 主分类:agent
  • 形态:method
  • 被引:6
  • 被引来源:Semantic Scholar
  • S2被引:6
  • OpenAlex被引:0
  • 影响力被引:0
  • TLDR:Single-rollout Asynchronous Optimization (SAO) is presented to address the stability and off-policy challenges in asynchronous RL and is able to train stably for one thousand steps and consistently outperform GRPO and its variants on agentic coding and reasoning benchmarks.
  • OpenAlex ID:W7167815111
  • OpenAlex DOI:10.48550/arxiv.2607.07508
  • DOI:10.48550/arxiv.2607.07508
  • DOI来源:OpenAlex
  • 开放获取:green
  • 开放获取链接:https://doi.org/10.48550/arxiv.2607.07508
  • OpenAlex更新:2026-07-19
  • 副分类:engineering
  • 待LLM分类:否
  • 标题中文:面向 Agentic 强化学习的单 Rollout 异步优化
  • TLDR中文:提出 Single-rollout Asynchronous Optimization(SAO),用于解决异步 RL 中的稳定性与 off-policy 难题,可稳定训练一千步,并在 Agentic 编码与推理基准上一致优于 GRPO 及其变体。
  • 来源文件
  • /inbox/tom/_candidates/2026-07-10-agent-rag-longcontext-candidates.json
  • [S2 enrich]
  • [OpenAlex backfill]