Group-SAE: Efficient Training of Sparse Autoencoders for Large Language Models via Layer Groups.

Davide Ghilardi, Federico Belotti, Marco Molinari, Tao Ma, Matteo Palmonari

Browse the full EMNLP paper archive.