Vending-Bench
收藏资源简介:
Vending-Bench是一个用于评估大型语言模型在商业决策中的长期一致性的模拟环境。该数据集通过模拟零售公司的运营,提供了动态的、逐月的管理模拟,让LLMs在每个月都根据前一个月的完整商业报告做出关键的战略决策。数据集的内容包括产品定价、订单规模、营销预算、招聘、解雇、贷款、培训费用、研发费用、销售预测、收入预测等。该数据集旨在评估LLMs在长期决策中的战略一致性、对市场变化的适应性以及决策的合理性。
Vending-Bench is a simulated environment for evaluating the long-term consistency of large language models (LLMs) in business decision-making. This dataset simulates the operations of retail enterprises, offering dynamic, month-over-month management simulations where LLMs make critical strategic decisions each month based on the complete business report from the preceding month. The dataset covers aspects including product pricing, order quantities, marketing budgets, hiring, employee layoffs, loan arrangements, training expenditures, R&D expenditures, sales forecasts, revenue forecasts, and more. This dataset is designed to evaluate the strategic consistency, adaptability to market changes, and decision-making rationality of LLMs in long-term business decision scenarios.




