Submissions should use full vocab size.

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

Submissions should use full vocab size. #472

Closed nv-rborkar closed 3 years ago

nv-rborkar commented 3 years ago

Submissions should not drop UNUSED tokens from vocabulary as this can cause a mathematical difference in how often MASK predictions hit as compared to that in reference implementation.

github-actions[bot] commented 3 years ago

MLCommons CLA bot All contributors have signed the MLCommons CLA ✍️ ✅