gradiend
Datasets
All datasets matching “gradiend”gradiend_race_data
GRADIEND Race Data
This dataset consists of templated sentences with the masked word being sensitive to race, e.g., African.
See GENTER and GRADIEND Religion Data for similar datasets.
Usage
The dataset uses one subset per class. Subset names are class identifiers: white, black, asian. Each subset has columns masked, split, and one column per class (e.g. white, black, asian) giving the token for that class in that row.
from datasets import load_dataset
# Load one… See the full description on the dataset page: https://huggingface.co/datasets/aieng-lab/gradiend_race_data.gradiend_religion_data
GRADIEND Religion Data
This dataset consists of templated sentences with the masked word being sensitive to religion, e.g., Jewish.
See GENTER and GRADIEND Race Data for similar datasets.
Usage
The dataset uses one subset per class. Subset names are class identifiers: jewish, christian, muslim. Each subset has columns masked, split, and one column per class (e.g. christian, jewish, muslim) giving the token for that class in that row.
from datasets import load_dataset… See the full description on the dataset page: https://huggingface.co/datasets/aieng-lab/gradiend_religion_data.gradiend_wiki_gender
WikiGender (GRADIEND – Wikipedia Gender)
WikiGender is a masked evaluation dataset derived from the English Wikipedia.It contains naturally occurring sentences with exactly one gendered target word, which is replaced by [MASK].
The dataset is designed to evaluate whether GRADIEND-style encoders and masked language models generalize gender-related representations to:
unseen lexical items
a different domain than the training source
Target Words
The following… See the full description on the dataset page: https://huggingface.co/datasets/aieng-lab/gradiend_wiki_gender.