Alibaba· QwenResearch paper

GSPO: Towards Scalable Reinforcement Learning for Language Models

Alibaba proposes the Group Sequence Policy Optimization (GSPO) algorithm to address instability and model collapse issues in existing reinforcement learning algorithms for language models, aiming to enable scalable RL training.

From the source

To enable successful RL scaling, we propose the Group Sequence Policy Optimization (GSPO) algorithm.

qwenlm.github.io

Announced

27 Jul 2025 07:00 UTC

Detected

03 Aug 2026 21:12 UTC (+372d after)