Hongmin chen

DASH: Deterministic Attention Scheduling for High-throughput Reproducible LLM Training

HSA: Head-wise Sparse Attention for Efficient and Accurate Long-context Inference