Presentation

DescriptionThe past few years have seen a surge of using Machine Learning (ML) and Deep Learning (DL) algorithms for traditional HPC tasks such as feature detection, numerical analysis, and graph analytics. While ML and DL enable solving HPC tasks, their adoption has been hampered due to the lack of understanding of how they utilize systems. Optimizing these algorithms requires characterizing their performance across the hardware/software (HW/SW) stack, but the lack of simple tools to automate the process and the reliance on researchers to perform manual characterization is a bottleneck. To alleviate this, we propose an across-stack profiling scheme and integrate it within MLModelScope — a hardware and software agnostic tool for evaluating and benchmarking ML/DL at scale. We demonstrate MLModelScope’s ability to characterize state-of-art ML/DL models and give insights that are only possible obtained by performing across-stack profiling.