Open jonathan-cohen-nvidia opened 4 years ago
General principle is that if you can, you should use pretrained weights provided by MLPerf (which should be hosted & indicated in the readme).
If you are unable to do so, you should submit anything necessary for an auditor to reproduce your pretrainining procedure.
For 0.7 I think we should just state the general principle with suggestions. For 0.8 we can make it more formal.
Submitted draft PR: https://github.com/mlperf/training_policies/pull/317
SWG:
This has been merged. We would love cleared rules going forward.
General principle is that if you can, you should use pretrained weights provided by MLPerf (which should be hosted & indicated in the readme).
If you are unable to do so, you should submit anything necessary for an auditor to reproduce your pretrainining procedure.
For 0.7 I think we should just state the general principle with suggestions. For 0.8 we can make it more formal.