Skip to content

rziga/min-xcodec2

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

3 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

min-xcodec2

"Minimal" implementation of xcodec2 with converted weights (available on huggingface). This is inference only. I have no intention of adding training support (at least not here). This is just meant to be copy-pastable to other projects. Verified to work up with tolerance of 5e-5 w.r.t. original (trust me bro).

Differences from original implementation:

  • Weights work without having to run any extra scripts.
  • Feature extraction for semantic encoder runs on the GPU so there is no syncs.
  • Removed most of the dead code.

Install with uv sync or pip install git+https://github.com/rziga/min-xcodec2

Run demo with uv run --group demo examples/demo.py.

Sources

License

MIT

About

"Minimal" implementation of xcodec2

Topics

Resources

License

Stars

Watchers

Forks

Contributors

Languages