SQUAD2

NAME = SQUAD2
DATASET_PATH = rajpurkar/squad_v2
SAMPLE_SPLIT = validation
FEWSHOT_SPLIT = train
RESPONSE_TYPE = COMPLETION
METRICS = [AccuracyCompletion, F1]
SUBJECTS = ['no_subject']
LANGUAGE = <Language.ENG: 'English'>

• Module: eval_framework.tasks.benchmarks.squad

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

• Link to dataset: https://huggingface.co/datasets/rajpurkar/squad_v2

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 "SQUAD2".