The idea that Keras would be the framework of choice for a research organization is laughable. If you're talking TF internals, those aren't really any more "acc

The idea that Keras would be the framework of choice for a research organization is laughable. If you're talking TF internals, those aren't really any more "accessible" than pytorch, which to many (including me) feels quite idiomatic