What is the meaning of "[1] Due to reference convergence issues, all BERT results normalize epochs to convergence for similar batch sizes" in ML-Perf v0.7 results.

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

Open yongdeok-kim opened 3 years ago

yongdeok-kim commented 3 years ago

Hi, I have a question on the some comments in MLPerf v0.7 training results. In the end of the table sheet, there is a sentence like

"[1] Due to reference convergence issues, all BERT results normalize epochs to convergence for similar batch sizes."

What is the exact meaning of it?