6 papers
Ring-Zero: Scaling Zero RL to a Trillion Parameters for Emergent Reasoning
Xinyu Tang, Qianggang Cao, Yurou Liu +13
The paper introduces a training pipeline that scales zero‑reinforcement‑learning to a trillion‑parameter language model, revealing emergent chain‑of‑thought reasoning abilities and…
MaP: A Unified Framework for Reliable Evaluation of Pre-training Dynamics
Jiapeng Wang, Changxin Tian, Kunlong Chen +5
Reliable evaluation is fundamental to the progress of Large Language Models (LLMs), yet the evaluation process during pre-training is plagued by significant instability that obscur…
MergeMix: Optimizing Mid-Training Data Mixtures via Learnable Model Merging
Jiapeng Wang, Changxin Tian, Kunlong Chen +5
Optimizing data mixtures is essential for unlocking the full potential of large language models (LLMs), yet identifying the optimal composition remains computationally prohibitive…
Rethinking Sample Polarity in Reinforcement Learning with Verifiable Rewards
Xinyu Tang, Yuliang Zhan, Zhixun Li +5
Large reasoning models (LRMs) are typically trained using reinforcement learning with verifiable reward (RLVR) to enhance their reasoning abilities. In this paradigm, policies are…
Towards High Data Efficiency in Reinforcement Learning with Verifiable Reward
Xinyu Tang, Zhenduo Zhang, Yurou Liu +4
Recent advances in large reasoning models have leveraged reinforcement learning with verifiable rewards (RLVR) to improve reasoning capabilities. However, scaling these methods typ…
WSM: Decay-Free Learning Rate Schedule via Checkpoint Merging for LLM Pre-training
Changxin Tian, Jiapeng Wang, Qian Zhao +7
Recent advances in learning rate (LR) scheduling have demonstrated the effectiveness of decay-free approaches that eliminate the traditional decay phase while maintaining competiti…