Raw experimental artefacts and summary figures for sparse-autoencoder
(SAE) feature-locality experiments across six base language models.
This README mirrors DATASET.md in the source repository
lccqqqqq/sae-analysis.
See that repository for the analysis pipelines that produce the
artefacts documented below.
This document describes the on-disk layout and per-file schemas for the
HuggingFace dataset lccqqqqq/sae-locality-data,
which mirrors the data/ tree… See the full description on the dataset page:
https://huggingface.co/datasets/Lccqqqqq/sae-locality-data.
