Porting AI/ML Models to Intelligence Processing Units (IPUs)
Porting AI/ML Models to Intelligence Processing Units (IPUs)
复制标题
DOI：
10.1145/3569951.3603632
复制
发表时间：
2023-07
期刊：
影响因子：
--
通讯作者：
Abhinand Nasari;Lujun Zhai;Zhenhua He;Hieu Hanh Le;S. Cui;Dhruva K. Chakravorty;Jian Tao;Honggao Liu
中科院分区：
文献类型：
--
作者：
Abhinand Nasari;Lujun Zhai;Zhenhua He;Hieu Hanh Le;S. Cui;Dhruva K. Chakravorty;Jian Tao;Honggao Liu
智能处理单元 (IPU) 是专门设计的加速器，专用于支持人工智能 (AI) 和机器学习 (ML) 工作流程。在此，我们报告了美国国家科学基金会 (NSF) 资助的新兴科学加速计算 (ACES) 测试平台上提供的 Graphcore IPU 的性能特征和代码移植经验。我们的基准测试将 ACES IPUS 上的 AI/ML 框架的性能与 Graphcloud 环境（Graphcore 提供的商业 IPU 云服务）上的类似运行进行了比较。我们还将两个 PyTorch 神经网络模型从图形处理单元 (GPU) 移植到 IPU，以确保软件环境的有效性。移植的模型包括用于从低分辨率图像重建高分辨率图像的 TransCycleGAN 模型，以及用于气候模型中大规模高分辨率科学数据压缩的分层自动编码器。使用 Graphcore Poplar 软件开发套件中的实用程序，这些模型已成功移植到多个 IPU 上。增加 IPU 数量会显着提高模型的吞吐量。
Intelligence processing units (IPUs) are specifically designed accelerators that are dedicated to support artificial intelligence (AI) and machine learning (ML) workflows. Here, we report on the performance characteristics and code-porting experiences on Graphcore IPUs offered on the new National Science Foundation (NSF)-funded Accelerating Computing for Emerging Sciences (ACES) testbed. Our benchmarks compared performance of AI/ML frameworks on ACES IPUS to similar runs on the Graphcloud environment, a commercial IPU cloud service offered by Graphcore. We also ported two PyTorch neural network models from Graphics Processing Units (GPUs) to IPUs to ensure the efficacy of the software environment. The ported models include the TransCycleGAN model that is used in reconstructing high-resolution images from low-resolution images, and the Hierarchical Autoencoder that is for large-scale high-resolution scientific data compression in climate models. These models were successfully ported on mulitple IPUs using utilities in the Graphcore Poplar software development kit. Increasing the number of IPUs resulted in a considerable enhancement in the model's throughput.