Was this with the `base` model? `large` is running ok on a P100 in colab, but is about 4% the speed of `base.en`. Certainly seems like some of these models will

Was this with the `base` model? `large` is running ok on a P100 in colab, but is about 4% the speed of `base.en`. Certainly seems like some of these models will be fast enough for real-time.