ARC_FI

NAME = ARC_FI
DATASET_PATH = LumiOpen/arc_challenge_mt
SAMPLE_SPLIT = test
FEWSHOT_SPLIT = validation
RESPONSE_TYPE = LOGLIKELIHOODS
METRICS = [AccuracyLoglikelihood, AccuracyNormLoglikelihood]
SUBJECTS = ['fi']
LANGUAGE = <Language.FIN: 'Finnish'>

• Module: eval_framework.tasks.benchmarks.arc_fi

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

• Link to dataset: https://huggingface.co/datasets/LumiOpen/arc_challenge_mt

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 "ARC_FI".