HumanEval

NAME = HumanEval
DATASET_PATH = openai/openai_humaneval
SAMPLE_SPLIT = test
FEWSHOT_SPLIT = test
RESPONSE_TYPE = COMPLETION
METRICS = [CodeCompletionAssertion]
SUBJECTS = ['no_subject']
LANGUAGE = <Language.ENG: 'English'>

• Module: eval_framework.tasks.benchmarks.humaneval

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

• Link to dataset: https://huggingface.co/datasets/openai/openai_humaneval

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 "HumanEval".