Open
Description
Currently the training API offers one preset configuration --train that offers a medium CPU to quality tradeoff. But there are many different parameters, which are non-trivial to understand. We should offer --train-fast and --train-best presets that offer the fastest training, and the highest quality training.