Closed galacticglum closed 4 years ago
I'm gonna close this issue for now since commit cac8ae1 has handled it; however, it may need later revising for a full Keras-compatible overhaul.
Turns out that a full Keras port was necessary for the Transformer model to integrate well with the rest of the project and CLI (which was already setup with Keras in mind). Commit 51206ea addresses this.
Since we are using Keras for the MusicRNN implementation, we may as well stay consistent across all models and use Keras everywhere. This will make the model usage API more streamlined and straightforward.