Last released Aug 8, 2026
Open-source toolkit for training and causally validating Sparse Autoencoder (SAE) features in Arabic LLMs using ALDi as external ground truth.
Supported by
