Add MLPerf logging + amendments

facebookresearch / dlrm

An implementation of a deep learning recommendation model (DLRM)

MIT License

3.72k stars 825 forks source link

Closed janekl closed 1 year ago

janekl commented 1 year ago

I'm adding MLPerf logging in this PR + several minor changes.

I made a few the decisions including:

removing test set from evaluation - because it's not used in MLPerf
skipping epoch end evaluation - this is done within epoch anyway (the last evaluation step during an epoch).

This is to make training loop and logging adequate for MLPerf purposes.

Thanks to @mmarcinkiewicz for helping out with all this.

janekl commented 1 year ago

MLPerf logging was integrated in the official reference implementation in https://github.com/mlcommons/training/pull/616. I'm closing the PR initially filed here.