datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
SAEVerbalizer-Data
SAEVerbalizer Data
Evaluation data for SAEVerbalizer: Generating Explanations for Sparse
Autoencoder Features via Representation Verbalization.
This release contains the paper's three evaluation sets for both the 27B
verbalizer and the 1B-to-27B adapter-based verbalization system. Training data
may be added in a later release.
Inference and Reference Agreement evaluation code is available in
THU-KEG/SAEVerbalizer.
Data collections
Collection
Feature space… See the full description on the dataset page: https://huggingface.co/datasets/THU-KEG/SAEVerbalizer-Data.llavasae_obliec100k_SAEV