AMSbench
收藏资源简介:
AMSbench是一个全面的数据集,旨在评估多模态大型语言模型(MLLMs)在模拟/混合信号(AMS)电路设计领域的感知、分析和设计能力。该数据集包含约8000个测试问题,跨越多个难度级别,并评估了八种主流模型。AMSbench的创建过程涉及收集来自学术界和工业界的多种数据模态,包括教科书、幻灯片、GitHub代码、电路图、表格/图表、行业报告和学术研究论文。数据集的构建旨在解决现有MLLMs在复杂多模态推理和高层次电路设计任务中的局限性,以推动自动化AMS电路设计工作流程的发展。
AMSbench is a comprehensive dataset designed to evaluate the perception, analysis, and design capabilities of multimodal large language models (MLLMs) in the analog/mixed-signal (AMS) circuit design domain. This dataset contains approximately 8,000 test questions spanning multiple difficulty levels, and has been utilized to assess eight mainstream models. The development of AMSbench involved gathering diverse data modalities from both academia and industry, including textbooks, lecture slides, GitHub code, circuit diagrams, tables, charts, industry reports, and academic research papers. The dataset is constructed to address the limitations of existing MLLMs in complex multimodal reasoning and high-level circuit design tasks, thereby promoting the development of automated AMS circuit design workflows.

- 1通过上海交通大学, 加州大学洛杉矶分校, 清华大学, 宁波东方理工学院 · 2025年



