DrafterBench
收藏资源简介:
DrafterBench是一个开源的自动工具包,旨在评估大型语言模型(LLM)代理在土木工程绘图修订任务中的自动化能力。该数据集包含了从现实世界绘图文件中总结的12种任务类型,共有46个定制化功能和1920个任务。DrafterBench通过模拟实际场景,测试AI代理在解读复杂和长上下文指令、利用先验知识和适应动态指令质量方面的能力。数据集通过系统分析任务准确性和错误统计数据,为集成LLM在工程应用中提供了深入见解。
DrafterBench is an open-source automated toolkit designed to evaluate the automation capabilities of large language model (LLM) agents in civil engineering drawing revision tasks. This dataset encompasses 12 task types summarized from real-world drawing files, along with 46 customized functions and 1920 individual tasks. DrafterBench simulates real-world scenarios to test AI Agents' abilities in interpreting complex and long-context instructions, leveraging prior knowledge, and adapting to dynamic instruction quality. By systematically analyzing task accuracy and error statistics, this dataset provides in-depth insights for the integration of LLMs in engineering applications.

- 1DrafterBench: Benchmarking Large Language Models for Tasks Automation in Civil Engineering麦吉尔大学土木工程系 · 2025年



