作者位次 #3/4ACL 2026Poster· Natural Language Processing
T ensor L ens: End-to-End Transformer Analysis via High-Order Attention Tensors
Ido Andrew Atad · Itamar Zimerman · Shahar Katz · Lior Wolf
摘要
Attention matrices are fundamental to transformer research, supporting a broad range of applications including interpretability, visualization, manipulation, and distillation. Yet, most existing analyses focus on individual attention heads or layers, failing to account for the model’s global beha...