Showcasing a new visualizer of MLPerf inference results and asking for feedback

[Grigori Fursin]

Dear all,

Our MLCommons taskforce has developed a prototype to visualize and compare private and public MLPerf inference results: https://cknowledge.org/cm-gui-graph .

We will demonstrate how to run it from command line using the MLCommons CM (CK2) framework to visualize your own results that you are preparing for the submission with all past public MLPerf inference results during our weekly conf-call tomorrow at 10:30am PST tomorrow (Thursday):  https://meet.google.com/jtk-zzbb-apx .

Please feel free to get in touch with me and Arjun with your feedback and feature requests that we can try to quickly implement to help you simplify your MLPerf inference submissions. You can also contact us via our public Discord channel.

For example, we were already asked to add CM extensions to visualize and compare ratios of standard MLPerf characteristics such as performance/Watt or performance per core/accelerator/rack or how SUT performance varies across different inference rounds, etc...

Looking forward to your feedback,
Grigori and Arjun