Activation Concentration: Characterizing Column-Level Output Sparsity Across Diffusion Model Architectures

Dazhi Yang, Shafayat Mowla Anik, Byeong Kil Lee +1

Recent diffusion accelerators exploit activation sparsity by skipping near-zero GELU outputs, reporting 52--85% element-level sparsity. However, systolic-array hardware processes a…