For what it's worth, we are pretty proud of the performance as well - I wouldn't call it training wheels :)
On both scale up and run times, it measures up as among the best-in-class[1]. That is to say, for the scenarios which people use it most commonly (GBT, linear learners), it's a great fit!
On both scale up and run times, it measures up as among the best-in-class[1]. That is to say, for the scenarios which people use it most commonly (GBT, linear learners), it's a great fit!
[1] https://arxiv.org/pdf/1905.05715.pdf