Qwenは、超長期（XLong）エージェント訓練にエラスティックRLを導入する基盤QwenGyreを発表した。Elastic SchedulerがrolloutとtrainingのGPUを需要に応じて再配分し、Colocate比で最大1.85×、Async比で最大1.78×高速化する。Qwen3.8-2.4T-A95Bに適用し、1回のrolloutあたり約700Kトークンの規模でNL2RepoBenchを52.5%から58.5%へ48ステップで改善、学習時間をAsync比で約44%削減した。
Introducing QwenGyre, bringing elastic RL to extreme-long-horizon (XLong) agent training. ✨
It launches with two core designs: 🥳
- Elastic Scheduler: reallocates GPUs between rollout & training as demand shifts—live executions never interrupted. Up to 1.85× speedup over Colocate, 1.78× over Async.
- Trajectory Processor: turns branching executions (compaction, sub-agents, retries) into prefix-shared trajectory trees with bounded, deduplicated training samples. Partial progress after timeout still gets scored.
Scaled to Qwen3.8-2.4T-A95B with ~700K tokens per rollout, QwenGyre lifts NL2RepoBench from 52.5% to 58.5% in 48 steps, cutting end-to-end training time by ~44% vs Async.
🔗 Read the paper:
https://arxiv.org/abs/2609.33848
出典: Qwen · x.com