Qingyu Yin

Currently: JDY Intern JoyAI, JD.com

Previously: Ace Toptalent Intern Hi Lab, Rednote (Xiaohongshu)

Scaling Foundation Model for OSS

Huggingface Google Scholar GitHub X (Twitter) Zhihu Email

Click here to see his publication list.

StableMask: Refining Causal Masking in Decoder-only TransformerICML'24

Qingyu Yin, Xuzheng He, Zhuang Zhuang, Yu Zhao, Jianhua Yao, Xiaoyu Shen, Qiang Zhang.

[ArXiv] [Code]

Constrain Alignment with Sparse AutoencodersICML'25

Qingyu Yin*, Chak Tou Leong*, Hongbo Zhang, Minjun Zhu, Hanqi Yan, Qiang Zhang, Yulan He, Wenjie Li, Jun Wang, Yue Zhang, Linyi Yang

Evaluating Parameter Efficient Methods for RLVRPreprint

Qingyu Yin*, Yulun Wu*, Zhennan Shen*, Sunbowen Li, Zhilin Wang, Yanshu Li, Chak Tou Leong, Jiale Kang, Jinjin Gu

[ArXiv] [Code]

Deeper Insights Without Updates: The Power of In-Context Learning Over Fine-TuningEMNLP'24 Findings

Qingyu Yin, Xuzheng He, Luoao Deng, Chak Tou Leong, Fan Wang, Yanzhao Yan, Xiaoyu Shen, Qiang Zhang.

[ArXiv] [Code]

ModRWKV: Transformer Multimodality in Linear TimeEMNLP'25 Oral

Jiale Kang*, Ziyin Yue*, Qingyu Yin*, Jiang Rui, Weile Li, Zening Lu, Zhouran Ji.

Refusal Falls off a Cliff: How Safety Alignment Fails in Reasoning?Preprint'25

Qingyu Yin*, Chak Tou Leong*, Linyi Yang, Wenxuan Huang, Wenjie Li, Xiting Wang, Jaehong Yoon, Jinjin Gu.

[PDF] [Code] [BibTeX]

Balancing LoRA Performance and Efficiency with Simple Shard SharingICLR'26

Jiale Kang, Qingyu Yin.

Open Source

An evaluation, environment, and production data tool for reinforcement learning agents.

A lightweight codebase and experimental framework for applying LoRA to RL algorithms.

[GitHub Repository]

An experimental simulation environment combining physics constraints with agent learning.

[GitHub Repository]