MetRex
收藏资源简介:
MetRex数据集由布朗大学工程学院创建,包含25,868个Verilog HDL设计及其对应的合成后指标,如面积、延迟和静态功耗。数据集内容丰富,涵盖了从简单到复杂的多种设计,旨在评估大型语言模型(LLMs)在硬件设计任务中的推理能力。数据集的创建过程包括自动化的数据收集、清洗和合成,确保了数据的高质量和可重复性。MetRex数据集主要应用于硬件设计领域,旨在解决LLMs在Verilog代码合成后指标推理中的挑战,提升模型在硬件设计中的实用性和准确性。
The MetRex dataset was developed by the School of Engineering at Brown University, consisting of 25,868 Verilog HDL designs and their corresponding post-synthesis metrics including area, latency, and static power consumption. The dataset covers a diverse range of designs spanning from simple to complex configurations, aiming to evaluate the reasoning capabilities of Large Language Models (LLMs) in hardware design tasks. The dataset creation process includes automated data collection, cleaning, and synthesis procedures, ensuring high data quality and reproducibility. Primarily applied in the field of hardware design, the MetRex dataset is designed to address the challenges of LLMs in reasoning about post-synthesis metrics of Verilog code, and improve the practicality and accuracy of models in hardware design applications.




