Fetching the paper…

DiTFastAttnV2: Head-wise Attention Compression for Multi-Modality Diffusion Transformers · Around