metadata
library_name: saelens
SAEs for use with the SAELens library
This repository contains the following SAEs:
- blocks.1.hook_resid_post
Load these SAEs using SAELens as below:
from sae_lens import SAE
sae, cfg_dict, sparsity = SAE.from_pretrained("chanind/sae-gemma-2-2b-layer-1-res-jumprelu", "<sae_id>")
This SAE has L0=80, and 0.93 variance explained.