EvalPerf Dataset

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)