Qwen releases QwenGyre, an elastic RL framework for long-horizon agent training

AISummary

Qwen introduces QwenGyre, an elastic reinforcement learning system for extreme-long-horizon agent training, with an Elastic Scheduler that reallocates GPUs between rollout and training.

The system reports speedups of up to 1.85x over Colocate and 1.78x over Async. Scaled to Qwen3.8-2.4T-A95B with about 700K tokens per rollout, QwenGyre lifts NL2RepoBench from 52.5% to 58.5% in 48 steps, cutting end-to-end training time about 44% versus Async.

Post on XView on X

Qwen

@Alibaba_Qwen

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

Source: Qwen · x.comPublished Oct 12, 9:11 AM ET · added here 9:42 AM ET

Qwen releases QwenGyre, an elastic RL framework for long-horizon agent training

Topics