metadata
library_name: saelens
SAEs for use with the SAELens library
This repository contains the following SAEs:
- 1000001536
- 833335296
- 333336576
- 166670336
- 500002816
- 666669056
Load these SAEs using SAELens as below:
from sae_lens import SAE
sae, cfg_dict, sparsity = SAE.from_pretrained("demiant/sae-gemma-2-2b-multistage-tied-unfrozen-5x-l1-jump-ortho-l1-5", "<sae_id>")