The aim of this project is to automate reproducible AI/ML performance benchmarking on a full stack of Red Hat platform. The benchmarkings of interest are the matrix multiply cuda code mini-benchmark and the MLPerf, which is a suite of fair and useful benchmarks that tests performance of ML software, hardware and services. In this project, mini-benchmarking and the MLPerf inference benchmarking are containerized and run in an automated fashion in Tekton CI/CD pipelines on Openshift 3.11 and higher. These benchmarking pipelines are also designed to run on CPUs and GPUs. The AI/ML workloads developed in this project will then be used for the larger full stack testing project which includes cross functional teams at Red Hat.
Red Hat Intern: Selbi Nuryyeva