Truss for Metavoice.
- Follows the server.py convention from the metavoice repo
flash_attnrequires installation with--no-build-isolation. As this isn't supported, installing the wheel directly seems to work.- Invoke with
truss predict -d '{"text": "This is a metavoice test"}' | python process.py(similar to Bark Truss example) – be aware of current 220 character limit on Metavoice
- Add support for picking the reference voice to use (e.g. see Whisper Truss example)
- Clean up config management (right now mostly copy-pasted from server.py)
- Accept longer text and split it to fit within the model limit
- Cache weights
- Move out process.py to a separate directory (or have model upload the file to a remote server)
- Get model to run on A10