BalancedCOPA

NAME = BalancedCOPA
DATASET_PATH = pkavumba/balanced-copa
SAMPLE_SPLIT = test
FEWSHOT_SPLIT = validation
RESPONSE_TYPE = LOGLIKELIHOODS
METRICS = [AccuracyLoglikelihood, AccuracyNormLoglikelihood]
SUBJECTS = ['no_subject']
LANGUAGE = <Language.ENG: 'English'>

• Module: eval_framework.tasks.benchmarks.balancedcopa

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

• Link to dataset: https://huggingface.co/datasets/pkavumba/balanced-copa

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 "BalancedCOPA".