GnomAD v3 random forest performance

In the GnomAD v3 blog post, the random forest for ancestry was evaluated using 20% of the samples as an evaluation set (gnomAD v3.1 New Content, Methods, Annotations, and Data Availability | gnomAD browser). Was the performance of the model published? I can’t seem to find it’s performance being evaluated anywhere.

Thanks,
Aidan