WINOX_DE

NAME = WINOX_DE
DATASET_PATH = demelin/wino_x
SAMPLE_SPLIT = test
FEWSHOT_SPLIT = test
RESPONSE_TYPE = LOGLIKELIHOODS
METRICS = [AccuracyLoglikelihood, AccuracyNormLoglikelihood]
SUBJECTS = ['lm_en_de']
LANGUAGE = <Language.DEU: 'German'>

• Module: eval_framework.tasks.benchmarks.winox

• File: src/eval_framework/tasks/benchmarks/winox.py | View on GitHub

• Link to dataset: https://huggingface.co/datasets/demelin/wino_x

More detailed documentation, with prompt examples and ground truth completions, can be generated with uv run -m eval_framework.utils.generate_task_docs --add-prompt-examples --only-tasks "WINOX_DE".