Jun Zhao

Why Reinforcement Fine-Tuning Enables MLLMs Preserve Prior Knowledge Better: A Data Perspective

Understanding Parametric and Contextual Knowledge Reconciliation within Large Language Models

Hardware-aligned Hierarchical Sparse Attention for Efficient Long-term Memory Access

Effective Length Extrapolation via Dimension-Wise Positional Embeddings Manipulation

Efficient Length-Generalizable Attention via Causal Retrieval for Long-Context Language Modeling