taylorbelrose

An implementation of Performer, a linear attention-based transformer, in Pytorch

Sparsify transformers with SAEs and transcoders

Tools for understanding how transformer predictions are built layer-by-layer