The EvalPerf dataset includes 121 performance-exercising tasks:
- task_id(- str): The task ID mapped to the original HumanEval(+) and MBPP(+) tasks
- reference(- List[str]): A list of representative reference solutions (from slow to fast) with diverse performance patterns
- pe_input(- Any): A performance exercising input
- scores(- List[float]): Performance scores for- reference(0-100)