Training Working Group

Mission

Define, develop, and conduct the MLPerf™ Training benchmarks.

Purpose

Benchmarking the performance of training ML models on a wide variety of use cases, software, and hardware drives AI performance across the tech industry. This working group draws on expertise in AI and the technology that powers AI from across the industry to design and create industry-standard benchmarks. Together, we create the reference implementations, rules, policies and procedures to benchmark a wide variety of AI workloads.

The Training working group strives for a critical balance of perspectives to ensure fairness and accuracy in the benchmarks. This balance comes from our member's diverse experience in many different AI hardware and software spaces. We are always looking for new members to help us create the benchmarks that best capture innovation in AI.

Deliverables

1. Training benchmark roadmap
2. Training benchmark rules
3. Training benchmark reference implementations
4. Training benchmark results every ~6 months

Meeting Schedule

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

How to Join and Access Working Group Resources

This group is limited to exclusively Members and Affiliates. If you are not already a Member/Affiliate or part of a Member/Affiliate company, you can learn more about Membership here.

• To sign up for the group mailing list, receive the meeting invite, and access shared documents and meeting minutes:
  1. Associate a Google account with your organizational email address.
  2. Request to join the Training Google Group. Requests are manually reviewed, so please be patient.
  3. Once your request to join the Training Google Group is approved, you'll be able to access the Training folder in the Members Google Drive.
• To engage in group dicussions:
  1. Join the group's channels on the MLCommons Discord server.
• To access the GitHub repository (public):
  1. If you want to contribute code, please sign our CLA first.
  2. Visit the GitHub repository.

Working Group Chairs

To contact all Training working group chairs email training-chairs@mlcommons.org.

Eric Han (erichan@mlcommons.org) - LinkedIn

Eric is a software engineer at Meta driving PyTorch performance improvements.

Ritika Borkar (ritika@mlcommons.org) - LinkedIn

Ritika Borkar is a Senior Deep Learning Architect at NVIDIA focusing on HW and SW optimizations for High Performance AI Computing on GPUs and datacenter systems. Previously, she worked on microarchitecture definition, ASIC design, and verification for IPs at Atmel and NVIDIA. Since MLPerf's inception in 2018, Ritika has influenced rules and processes for the training suite of benchmarks. She holds a master's degree in Electrical Engineering from the University of Minnesota, and a bachelor's degree from the National Institute of Technology at Trichy in India.