search

mlcommons / training_policies

Issues related to MLPerf™ training policies, including rules and suggested changes
https://mlcommons.org/en/groups/training
Apache License 2.0
92 stars 66 forks source link

NV DLRM update #407

Closed johntran-nv closed 3 years ago

johntran-nv commented 3 years ago

Slides for today's presentation are here: https://docs.google.com/presentation/d/1o-p_-eooTdjQfedNFYV_o1GO9VK0bys72hro78el6kg/edit?usp=sharing

PR is here: https://github.com/facebookresearch/dlrm/pull/154

johntran-nv commented 3 years ago

From this week's meeting, everyone generally agreed to change the reference to match the best numerical behavior, and to allow submitters to have worse behavior (and worse convergence) if they choose to. Waiting for PR from above to go through.

johntran-nv commented 3 years ago

I believe this has been resolved for a while now. Closing.