from dataclasses import dataclass from enum import Enum @dataclass class Task: benchmark: str metric: str col_name: str class Tasks(Enum): # Core HSO & Hallucination Potential Metrics xray_score = Task("xray_score", "score", "X-Ray Risk ⬇️") struct_risk = Task("structural_risk", "score", "Structural Risk ⬇️") behav_risk = Task("behavioral_risk", "score", "Behavioral Risk ⬇️") stable_rank = Task("stable_rank", "score", "Stable Rank (r_s) ⬆️") effective_rank = Task("effective_rank", "score", "Effective Rank (r_eff) ⬆️") cond_num = Task("condition_number", "score", "Condition (κ) ⬇️") blind_fraction = Task("blind_fraction", "score", "Blind Manifold (ker %) ⬇️") factuality = Task("factuality", "score", "Factuality (%) ⬆️") paraphrase_stab = Task("paraphrase_stability", "score", "Paraphrase Stability (%) ⬆️") # Standard Benchmark Overlays overall_mixeval = Task("MixEvalScorer.overall", "score", "MixEval Overall ⬆️") gsm8k = Task("MixEvalScorer.GSM8k", "score", "GSM8k ⬆️") math = Task("MixEvalScorer.MATH", "score", "MATH ⬆️") mmlu = Task("MixEvalScorer.MMLU", "score", "MMLU ⬆️") gpqa = Task("MixEvalScorer.GPQA", "score", "GPQA ⬆️") arc = Task("MixEvalScorer.ARC", "score", "ARC ⬆️") NUM_FEWSHOT = 0 TITLE = """