Fetching the paper…

AttnLRP: Attention-Aware Layer-Wise Relevance Propagation for Transformers · Around