This repository derived by modifying original PianoTree VAE model and code.
$ pip install numpy torch tqdm tensorboardx matplotlib
Adapt train.py to work with 32 or 128 time steps.
model.py is the original PianoTree VAE that uses GRU blocks. model_tr_full_ultra.py is the full transformer VAE model designed for 32 time steps. model_tr_full_ultra_plus.py adapted version to work with 128 time steps.
Download weights from here, and put them in ./runs folder.
See test.ipynb for 32 steps model inference, and test_128.ipynb for 128 respectively.
