IFEval

NAME = IFEval
DATASET_PATH = google/IFEval
SAMPLE_SPLIT = train
FEWSHOT_SPLIT = train
RESPONSE_TYPE = COMPLETION
METRICS = [IFEvalMetric]
SUBJECTS = ['no_subject']
LANGUAGE = {'no_subject': <Language.ENG: 'English'>}

• Module: eval_framework.tasks.benchmarks.ifeval

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

• Link to dataset: https://huggingface.co/datasets/google/IFEval

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 "IFEval".