1 paper
Daewon Chae, Hyunwon Chung, Changwoo Lee +1
Large-scale foundation models achieve strong performance across diverse tasks, but their size makes inference costly, largely due to dense matrix multiplications. Prior work reduce…