Xiaoying Jia

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

Model Merging in Pre-training of Large Language Models