Tags: openmirlab/mt3-infer

Release 0.2.0

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01J1meSbZ1oPTneyYvEYnV4a

feat: add Colab notebook and remove note_seq dependency

- Add notebooks/quickstart_colab.ipynb following melband-roformer-infer style
- Remove note_seq dependency completely (was causing protobuf<4.0 conflicts)
- Add lightweight NoteSequence implementation (note_sequence_lite.py)
- Add MIDI constants module to replace note_seq constants
- Update mt3_pytorch adapter to use pretty_midi for MIDI conversion
- Fix transformers 4.44+ compatibility (add cache_position to T5Block calls)
- Fix checkpoint import for newer transformers (use torch.utils.checkpoint)
- Relax transformers from ~=4.30.2 to >=4.35.0
- Relax torch/torchvision/torchaudio from ==2.7.1 to >=2.0.0
- Update check_torch_version() to allow torch>=2.0.0
- Update requires-python from >=3.8 to >=3.9
- Bump version to 0.1.3

This makes mt3-infer fully compatible with Google Colab (Python 3.12,
torch 2.9.0, transformers 4.57.5, protobuf 5.x) without any dependency conflicts.

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Remove all training-related code and bump version to 0.1.2

- Removed wandb dependency requirement
- Removed training methods from YourMT3 (training_step, validation_step, etc.)
- Removed training-specific files (init_train.py, optimizers.py, lr_scheduler.py, etc.)
- Removed unused T5Adversarial class from MR-MT3
- Cleaned up training-related parameters from YourMT3 __init__
- Verified MR-MT3 and MT3-PyTorch are inference-only
- Updated version to 0.1.2

Fix: Include YAML config files in package distribution

Prepare v0.1.0 for first PyPI release

- Set version to 0.1.0 in pyproject.toml (first PyPI publish)
- Update CHANGELOG.md date for v0.1.0 to 2025-11-14
- Fix version consistency in README.md
- Add GitHub Actions workflow for automatic PyPI publishing