发表机构
Distyl AI; Polytechnique Montréal; Mila – Quebec AI Institute（Distyl AI; 蒙特利尔理工学院; 米拉-魁北克人工智能研究所）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
HARDEN通过约束进化搜索在保持答案不变的前提下生成更难的评测用例，在多个基准和模型上平均降低准确率22.7%，最高达49.9%。
AI 中文摘要
语言模型通常在精心策划的基准上进行评估，而这些基准未能充分代表企业部署的复杂性。我们引入了HARDEN，一种约束进化搜索方法，用于将现有评测用例的输入调整为更具挑战性的变体，同时保持其预期输出不变。HARDEN沿着生成的领域特定复杂性轴进行搜索，同时强制执行可行性约束，如保持任务语义、真实性和执行有效性。在FinQA、PubMedQA和ContractNLI以及三种Qwen3.5模型规模（35B-A3B、122B-A10B和397B-A17B）上，HARDEN将任务-模型准确率平均降低了22.7%，相对于使用相同可行性检查的单遍基线，最高降低了49.9%。这些结果表明，进化搜索可以产生明显更难的有效评测用例。
英文摘要
Language models are often evaluated on curated benchmarks that underrepresent the complexity of enterprise deployments. We introduce HARDEN, a constrained evolutionary search method to adapt the input of existing evaluation cases into more challenging variants while keeping their expected outputs fixed. HARDEN searches along generated domain-specific complexity axes while enforcing feasibility constraints such as preserving task semantics, realism, and execution validity. Across FinQA, PubMedQA, and ContractNLI and three Qwen3.5 model scales (35B-A3B, 122B-A10B, and 397B-A17B), HARDEN reduces task-model accuracy by 22.7% on average and by up to 49.9% relative to single-pass baselines using the same feasibility checks. These results show that evolutionary search can produce substantially harder valid evaluation cases.
