Red Hat is excited to announce that we have joined MLCommons, an open engineering consortium that curates the MLPerf benchmarking suite, as a founding member. MLCommons will be focusing on three important pillars to support the artificial intelligence (AI) and machine learning (ML) community: benchmarks, data sets, and best practices.
MLPerf is an established tool used to evaluate software frameworks, hardware platforms, and cloud platforms for AI and ML performance. The MLPerf benchmark suite represents the major application areas of AI and continues to evolve, adding new benchmarks that facilitate state-of-the-art innovation across different market segments.
MLCommons takes an open source-centric approach to support a variety of AI and ML environments, including mobile, high performance computing (HPC) and cloud computing, and is focused on the development of:
Datasets for training ML models
Benchmark result publications
Red Hat, the world’s leading provider of open source solutions, has extensive experience developing specifications, and optimizing and running a variety of benchmarks across multiple industry-standard benchmarking organizations. Red Hat is helping to found MLCommons because we believe that the MLPerf benchmark suite will benefit from a more formal collaborative effort to establish a standard and become a repository for data sets and best practices for AI and ML performance.
Red Hat and MLPerf
Red Hat has made an early investment in identifying and consolidating common AI workflows by forming the AI Center of Excellence, which helps integrate AI strategy across the Red Hat portfolio. It is through those efforts that we became involved in MLPerf, making it the de facto tool for measuring the AI and ML performance of our software. We have since employed the MLPerf benchmark suite to conduct full stack testing, create several reference architectures with our hardware partners, and work on AI and ML enablement and end-to-end validation of our software stack.
Per this involvement, Red Hat became a contributor to MLPerf via multiple working groups, such as the best practices group focused on MLCube, a set of common conventions for creating ML software that can "plug-and-play" various systems. MLCube is designed to make it easier for researchers to share innovative ML models, developers to experiment with different models and software companies to create infrastructure for models.
We envision a growing demand for benchmark results from the academic community, customers and hardware manufacturers alike. The best practices working group facilitates efforts to enhance the MLCube user experience and promote the adoption of MLPerf benchmarks by enhancing usability, interoperability and reliability of ML models, including running benchmarks in containerized infrastructures, such as Red Hat OpenShift.
Red Hat is proud to contribute to these efforts as a founding member of MLCommons. We look forward to working together to advance ML markets, democratize knowledge and deliver a set of comparable and relevant results to global communities.
Learn more about MLCommons
About the author
Daniel Riek is responsible for driving the technology strategy and facilitating the adoption of Analytics, Machine Learning, and Artificial Intelligence across Red Hat. Focus areas are OpenShift / Kubernetes as a platform for AI, application of AI development and quality process, AI enhanced Operations, enablement for Intelligent Apps.