Rui-Jie Zhu

Reformulation for Pretraining Data Augmentation

Scaling Linear Attention Capacity with Sparse State Expansion

Energy Efficient Language Models through Dynamic Sparsity

ZeCO: Zero-Communication Overhead Sequence Parallelism for Linear Attention

Quantized Spike-driven Transformer

CopyLens: Dynamically Flagging Copyrighted Sub-Dataset Contributions to LLM Outputs