Trainable Log-linear Sparse Attention for Efficient Diffusion Transformers

Re-assign community