用高效框架训练出性能更强的代码智能体，成本更低。
SkyRL-Agent: Efficient RL Training for Multi-turn LLM Agent
- 异步调度+轻量工具集成，提升长时序智能体训练效率。
- 纯强化学习训练出32B代码智能体，通过率提升至39.4%。
- 可拓展至研究、计算机使用等多类任务，适配不同训练后端。
我们提出 SkyRL-Agent，一个用于高效多轮、长时程智能体训练与评估的框架。它提供高效的异步分发、轻量级工具集成和灵活的后端互操作性，可无缝接入 SkyRL-train、VeRL、Tinker 等现有强化学习框架。基于该框架，我们训练了 SA-SWE-32B——一个从 Qwen3-32B（Pass@1 为 24.4%）出发，完全通过强化学习训练的软件工程智能体。核心改进包括：优化的异步流水线调度器，相较朴素异步批处理提升 1.55 倍速度；以及基于抽象语法树（AST）的搜索工具增强训练方案，促进代码导航，提升采样通过率（Pass@K），并提高训练效率。联合优化使 SA-SWE-32B 在 SWE-Bench Verified 上达到 39.4% Pass@1，且训练成本降低超过 2 倍。尽管仅在 SWE 任务上训练，该模型仍能有效泛化至终端、浏览、网页交互等其他智能体任务。我们还通过深度研究、计算机使用、记忆智能体等案例验证了 SkyRL-Agent 的可扩展性，各任务使用不同训练后端完成。
原文摘要 · Abstract (English)
We introduce SkyRL-Agent, a framework for efficient, multi-turn, long-horizon agent training and evaluation. It provides efficient asynchronous dispatching, lightweight tool integration, and flexible backend interoperability, enabling seamless use with existing RL frameworks such as SkyRL-train, VeRL, and Tinker. Using SkyRL-Agent, we train SA-SWE-32B, a software engineering agent trained from Qwen3-32B (24.4% Pass@1) purely with reinforcement learning. We introduce two key components: an optimized asynchronous pipeline dispatcher that achieves a 1.55x speedup over naive asynchronous batching, and a tool-enhanced training recipe leveraging an AST-based search tool to facilitate code navigation, boost rollout Pass@K, and improve training efficiency. Together, these optimizations enable SA-SWE-32B to reach 39.4% Pass@1 on SWE-Bench Verified with more than 2x cost reduction compared to prior models reaching similar performance. Despite being trained solely on SWE tasks, SA-SWE-32B generalizes effectively to other agentic tasks, including Terminal-Bench, BrowseComp-Plus, and WebArena. We further demonstrate SkyRL-Agent's extensibility through case studies on deep research, computer use, and memory agents, each trained using a different training backend.
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。