Open/Preview submissions from Inference v2.1
In the MLPerf Inference v2.1 round, Qualcomm and their partners submitted a number of RetinaNet results, which the Review Committee eventually accepted to the Open division under the Preview category. Also, ResNet50 and BERT results from the same systems were accepted to the Closed division under the Available category.
Given the extenuating circumstances surrounding these RetinaNet submissions, I believe it would be reasonable NOT to require ALL of such submissions to be replayed in the v3.0 round under the Available category, for example, if a system is no longer available for experimentation.
Discussed two options:
- Submit in v3.0 and withdraw during review period so that preview results stay in v2.1 table.
- Not submitted in v3.0, and mark v2.1 preview results as retracted/invalid (exact action TBD based on the discussion from chairs meeting)
Option 2 was chosen
David is going to propose rule changes for Preview->Available
This issue has been open for 1.5 years. Let's close it?