v0.1.2

Checkpoints moved to Hugging Face

The original download links on storage.googleapis.com/esun-ai/g2pW/ are being taken down. Checkpoints are now hosted at YC-Chen/g2pW (byte-identical files):

• G2PWModel-v2.zip
• G2PWModel-v2-onnx.zip

G2PWConverter() now downloads the model from Hugging Face. Please upgrade (pip install -U g2pw): on 0.1.1 and earlier, the automatic model download will fail once the GCS links are gone.

Changes

• Download model from Hugging Face instead of GCS (#35)
• Add requests to install_requires (transformers>=5 no longer pulls it in)
• Fix G2PW.from_pretrained on transformers>=5 and loading old checkpoints on newer transformers (#35)
• Fix scripts/predict_g2p_bert.py, which was calling the ONNX predict (#35)
• Fix scripts/export_onnx_model.py on torch>=2.9 / recent transformers (#35)
• Optional GPU inference via G2PWConverter(use_cuda=True)
• Read config files as UTF-8 (#29)

Full Changelog: v0.1.1...v0.1.2