Comment by meghan_rain

I welcome all progress but I don't see why these models aren't simply run on a thin Python server that loads the model into memory once and then you can curl it instantly whenever you want?