List your company
ResearchFormerly WizWand

SAE latent interpretation on Llama Scope SAEs (test)

0.501Hit RateSelfIE (SA)
0.223320.295410.36750.43959Feb 10, 2026
Updated 8mo ago

Evaluation results

MethodLinks
SelfIE (SA)Training Source=Goodfire SAE, Method Type=Scalar Affine, Sampling strategy=best-of-62026.020.5010.6710.774
SelfIE (SA)Training Source=Llama Scope SAE, Method Type=Scalar Affine, Sampling strategy=best-of-62026.020.4920.6780.794
SelfIE (SA+LR)Training Source=Llama Scope SAE, Method Type=Scalar Affine + Rank-16, Sampling strategy=best-of-62026.020.4780.6860.814
Original + 5 ParaphrasesLabel candidates=6, Sampling strategy=best-of-62026.020.4470.6780.816
SelfIE (SA+LR)Training Source=Goodfire SAE, Method Type=Scalar Affine + Rank-16, Sampling strategy=best-of-62026.020.4450.6250.772
Auto-Interp Labels x6Label candidates=6, Sampling strategy=best-of-62026.020.4190.6550.821
SelfIE (SA)Training Source=Wikipedia topics, Method Type=Scalar Affine, Sampling strategy=best-of-62026.020.4110.5640.627
Untrained SelfIETraining Source=None, Sampling strategy=best-of-62026.020.3130.4880.684
SelfIE (Full-Rank)Training Source=Wikipedia topics, Method Type=Full-Rank, Sampling strategy=best-of-62026.020.2340.3610.512