RN50 needs an opt_name hyper parameter

mlcommons / training_policies

Issues related to MLPerf™ training policies, including rules and suggested changes

https://mlcommons.org/en/groups/training

Apache License 2.0

93 stars 66 forks source link

Closed jonathan-cohen-nvidia closed 4 years ago

jonathan-cohen-nvidia commented 4 years ago

Add to HP table. Allowed values are "sgd" or "lars" with no constraints.

Needed so that optimizer choice can be stolen post-submission.

bitfort commented 4 years ago

This is currently in the logging checker.

Table updated. Clarifying that optimizer is an HP choice that can be borrowed/stolen/copied during review.