Weight-based decomposition : a case for bilinear MLPs
Bron
ICML 2024 Workshop on Mechanistic Interpretability, July 27, 2024, Vienna, Austria- () p. 1-20
Tokenized SAEs : disentangling SAE reconstructions
Bron
ICML 2024 Workshop on Mechanistic Interpretability, July 27, 2024, Vienna, Austria- () p. 1-13