Inference Working Group

Mission

Create a set of fair and representative inference benchmarks.

Purpose

Machine-learning (ML) hardware and software system demand is burgeoning. Driven by ML applications, the number of different ML inference systems has exploded. Over 100 organizations are building ML inference chips, and the systems that incorporate existing models span at least three orders of magnitude in power consumption and five orders of magnitude in performance; they range from embedded devices to data-center solutions. Fueling the hardware are a dozen or more software frameworks and libraries. The myriad combinations of ML hardware and ML software make assessing ML-system performance in an architecture-neutral, representative, and reproducible manner challenging. There is a clear need for industry-wide standard ML benchmarking and evaluation criteria. MLPerf™ Inference answers that call.

Deliverables

1. Inference benchmark rules and definitions
2. Inference benchmark reference software
3. Inference benchmark submission rules
4. Inference benchmark roadmap
5. Publish inference benchmark results every ~6 months

Meeting Schedule

Weekly on Tuesday from 8:35-10:00AM Pacific.

How to Join

Use this link to request to join the group/mailing list, and receive the meeting invite:
Inference Google Group.
Requests are manually reviewed, so please be patient.

Working Group Resources

• Shared documents and meeting minutes:

  1. Associate a Google account with your e-mail address.
  2. Ask to join our Public Google Group.
  3. Ask to join our Members Google Group.
  4. Once approved, go to the Inference folder in the Members Google Drive.

• GitHub (public)

  1. If you want to contribute code, please sign our CLA first.
  2. GitHub link.

Working Group Chairs

Mitchelle Rasquinha (mrasquinha@mlcommons.org) - LinkedIn

Mitchelle Rasquinha is a Senior Software Engineer working on the ML Performance Team within Google. She is interested in accurately capturing innovations in system architectures through robust benchmarking. Mitchelle has a background in Computer Architecture from the Georgia Institute of Technology.

Miro Hodak (miro@mlcommons.org) - LinkedIn

Miro Hodak is a Senior Member of Technical Staff at AMD where he works on AI performance, strategy, and solutions. Before joining AMD, he worked as an AI Architect at Lenovo Infrastructure Solutions Group, and, prior to that, he was a Research Assistant Professor in Physics at North Carolina State University. Miro has participated in MLPerf/MLCommons activities since 2020 including submitting multiple rounds of Inference and Training benchmarks. Miro has journal publications spanning AI, computer science, materials science, physics, and biochemistry. His work has been cited over 2,000 times.