- Download the dataset from the Baroque-NMT repo
- Run
add_multipart_tracks.py
- Then run
encode.py
andtranspose.py
- Run
train.py
with arguments to customize the model architecture. - This should generate model checkpoints (in the
models
directory by default) - Run
inference.py
and provide a task and the path to the trained checkpoint