Reinforcement learning (RL) has emerged as an effective post-training paradigm for enhancing the reasoning capabilities of multimodal large language model (MLLM). However, current RL pipelines often suffer from training inefficiencies caused by two underexplored issues: Advantage Collapsing, where most advantages in a batch concentrate near zero, and Rollout Silencing, where the proportion of rollouts contributing non-zero gradients diminishes over time. These issues lead to suboptimal gradient updates and hinder long-term learning efficiency. To address these issues, we propose Shuffle-R1, a simple yet principled framework that improves RL fine-tuning efficiency by dynamically restructuring trajectory sampling and batch composition. It introduces (1) Pairwise Trajectory Sampling, which selects high-contrast trajectories with large advantages to improve gradient signal quality, and (2) Advantage-based Trajectory Shuffle, which increases exposure of valuable rollouts through informed batch reshuffling. Experiments across multiple reasoning benchmarks show that our framework consistently outperforms strong RL baselines with minimal overhead. These results highlight the importance of data-centric adaptations for more efficient RL training in MLLM.
@article{arxiv.2508.05612,
title = {Shuffle-R1: Efficient RL framework for Multimodal Large Language Models via Data-centric Dynamic Shuffle},
author = {Linghao Zhu and Yiran Guan and Dingkang Liang and Jianzhong Ju and Zhenbo Luo and Bin Qin and Jian Luan and Yuliang Liu and Xiang Bai},
journal= {arXiv preprint arXiv:2508.05612},
year = {2026}
}
Comments
This paper has been accepted by ICLR 2026. Conference link: https://iclr.cc/virtual/2026/poster/10007559 OpenReview link: https://openreview.net/forum?id=mYP33u1QBK Project page at: https://xenozlh.github.io/Shuffle-R1/