Kewei Tu

Flash Multi-Head Feed-Forward Network

Efficient Multimodal Planning Agent for Visual Question-Answering

Repurposing Synthetic Data for Fine-grained Search Agent Supervision

Scaling Agents via Continual Pre-training

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

Efficient Long-range Language Modeling with Self-supervised Causal Retrieval

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