Hacker News· helloericsf·· 6 天前AI 评分34
QwenGyre：面向超长程智能体训练的弹性 RL 框架
QwenGyre: An Elastic RL Framework for Training XLong-Horizon Agents
AI 导读
阿里 Qwen 团队提出 QwenGyre，一个面向超长程在线 RL 的端到端框架，可在不中断执行的情况下弹性重分配 GPU，并通过轨迹处理器重建分支历史、评估部分进展、去重冗余路径。
正文 · 原文
Authors:Weiqi Wang, Yuxin Zhou, Mouxiang Chen, Siyuan Zhang, Yi Zhang, Yuyan Luo, Zhiyu Yin, Chencan Wu, Jiemin Jiang, Wentao Yao, Chujie Zheng, JianWei Zhang
Abstract:Large language model (LLM) agents increasingly undertake extreme-long (xlong) horizon tasks, where a single execution can span hours, hundreds of model--environment interactions, and nearly 1M tokens per rollout. Applying online reinforcement learning (RL) to such executions poses two fundamental challenges: (1) severe execution variance and prolonged rollout delays cause massive GPU idling; and (2) complex non-linear branching generates massive trajectory redundancy, crippling training efficiency. To address these, we presents QwenGyre, an end-to-end framework for xlong-horizon online RL. QwenGyre elastically reallocates GPUs between rollout and training without interrupting live executions, while its trajectory processor reconstructs branching histories, scores partial progress, and deduplicates redundant paths to bound training costs. Scaled to our flagship model, Qwen~3.8 2.4T, with 700K tokens per rollout, QwenGyre yields a 6.0% absolute gain on NL2RepoBench (52.5% $\to$ 58.5%) in 48 steps. Across our evaluations on diverse domains of training datasets, QwenGyre delivers up to $1.85\times$ and $1.78\times$ speedups over Colocate and Async, respectively.
| Subjects: | Machine Learning (cs.LG); Distributed, Parallel, and Cluster Computing (cs.DC) |
| Cite as: | arXiv:2609.33848 [cs.LG] |
| (or arXiv:2609.33848v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2609.33848 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Mouxiang Chen [view email]
[v1]
Sun, 27 Sep 2026 19:02:03 UTC (535 KB)
来源：Hacker News · arxiv.org