Dependency Update, Training Improvements, and a PyPI Package

[JinruG]

Dependency Compatibility

The original codebase pinned dependencies to versions from 2018–2019 that are now incompatible with modern environments. This PR brings them up to date while preserving full backward compatibility with the existing model architecture and training logic:

Python ≥ 3.9
PyTorch ≥ 2.0.0
music21 ≥ 8.1.0
numpy ≥ 1.23.0

Training Improvements

• cuda() called before train() : The model previously was moved to GPU only after training, causing the entire training loop to run on CPU.
• Best-val-loss checkpointing : Each voice model now saves only the checkpoint that achieved the lowest validation loss, rather than always overwriting with the final epoch.
• ReduceLROnPlateau scheduler : Learning rate is automatically reduced when validation loss plateaus, controlled by the new --lr_patience and --lr_factor options above.

Dependency Compatibility

deepbach_pytorch PyPI Package

https://pypi.org/project/deepbach-pytorch/0.3.6/

A test script (test.py) is included that exercises the public API entry point:

test.py