Updating allowed optimizer table to match analysis

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

Updating allowed optimizer table to match analysis #319

Closed jonathan-cohen-nvidia closed 4 years ago

jonathan-cohen-nvidia commented 4 years ago

Added BERT, DLRM, split RN50 SGD from RN50 LARS, and added not that there are no compliant RN50 LARS implementations in PyTorch and no compliant RN50 SGD implementations in MxNet.

See: https://github.com/mlperf/training_policies/issues/318