EleutherAI / elk

Keeping language models honest by directly eliciting knowledge encoded in their activations.
MIT License
186 stars 33 forks source link

fix: remove first layer (embeddings) from layer_indices #284

Closed lauritowal closed 1 year ago

lauritowal commented 1 year ago

There was a bug where the first layer