Fetching the paper…

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