CausalGym is a benchmark for comparing the performance of causal interpretability methods
on a variety of simple linguistic tasks taken from the SyntaxGym evaluation set
(Gauthier et al., 2020, Hu et al., 2020)
and converted into a format suitable for interventional interpretability.
The dataset includes train/dev/test splits (exactly as used in the experiments in the paper).
The base/src columns are the prompts on which intervention is done. Each of these is a list of strings,
with each… See the full description on the dataset page:
https://huggingface.co/datasets/aryaman/causalgym.
