Koios
收藏资源简介:
Koios数据集是由德克萨斯大学奥斯汀分校创建的一套深度学习加速基准电路集,包含19个电路设计,覆盖多种神经网络类型、设计规模和数值精度。数据集旨在为FPGA架构和CAD研究提供更贴近实际的基准,通过高度的数据并行性、异构性和流水线深度,以及对FPGA架构特性的广泛利用,帮助研究人员识别架构低效并优化CAD工具。该数据集适用于探索针对深度学习优化的FPGA架构,解决现有学术基准集在深度学习领域不足的问题。
The Koios dataset is a deep learning acceleration benchmark circuit suite created by The University of Texas at Austin, which includes 19 circuit designs covering various neural network types, design scales and numerical precisions. This dataset aims to provide more practical benchmarks for FPGA architecture and CAD research. By leveraging high degrees of data parallelism, heterogeneity and pipeline depth, as well as extensive utilization of FPGA architectural characteristics, it helps researchers identify architecture inefficiencies and optimize CAD tools. This dataset is applicable to exploring deep learning-optimized FPGA architectures, addressing the shortcomings of existing academic benchmark suites in the deep learning field.




