Tango

Tango是一个专为评估各种加速器而设计的深度神经网络基准套件。它包含五种最广泛使用的卷积神经网络和两种循环神经网络，支持CUDA和OpenCL平台。该套件通过在架构模拟器、服务器GPU、移动GPU和移动FPGA上运行，提供了这些网络的深入架构统计。Tango的目标是为神经网络算法研究者和计算机架构师提供一个可广泛使用的评估环境，以便在硬件和软件层面上验证他们的想法。

Tango is a deep neural network benchmark suite specifically designed for evaluating various accelerators. It includes five most widely used convolutional neural networks (CNNs) and two recurrent neural networks (RNNs), and supports both CUDA and OpenCL platforms. This suite provides in-depth architectural statistics of these networks by running on architectural simulators, server GPUs, mobile GPUs and mobile FPGAs. The goal of Tango is to provide a widely usable evaluation environment for neural network algorithm researchers and computer architects, so that they can validate their ideas at both hardware and software levels.