Qwen은 초장기(XLong) 에이전트 학습을 위한 기반인 QwenGyre를 발표했다. 이는 탄력적 RL을 도입한 것이다. Elastic Scheduler가 rollout과 training의 GPU를 수요에 따라 재배분하여, Colocate 방식 대비 최대 1.85배, Async 방식 대비 최대 1.78배 빨라진다. Qwen3.8-2.4T-A95B에 적용했으며, rollout 1회당 약 700K 토큰 규모에서 NL2RepoBench 점수를 48스텝 만에 52.5%에서 58.5%로 개선했고, 학습 시간은 Async 방식 대비 약 44% 줄었다.
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
출처: Qwen · x.com