COPA_IDK

NAME = COPA_IDK
DATASET_PATH = aps/super_glue
SAMPLE_SPLIT = validation
FEWSHOT_SPLIT = test
RESPONSE_TYPE = LOGLIKELIHOODS
METRICS = [AccuracyLoglikelihood, AccuracyNormLoglikelihood, ConfidenceWeightedAccuracy, DistributionalCorrectnessScore, TernaryScore]
SUBJECTS = ['copa']
LANGUAGE = <Language.ENG: 'English'>

• Module: eval_framework.tasks.benchmarks.copa

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

• Link to dataset: https://huggingface.co/datasets/aps/super_glue

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 "COPA_IDK".