Hi Kyutai team, thanks for releasing Hibiki. I read through the README and CONTRIBUTING.md, and the project is already clearly connected to the paper, samples, Hugging Face models, Colab, and the different inference backends.
I think the README could still benefit from a small GitHub/open-source presentation pass to make the project easier to scan for new visitors. A few possible improvements:
- Add badges near the top, for example license, paper/arXiv, Hugging Face models, Colab, Python version, and supported backends.
- Make the mixed licensing easier to see at a glance. The repo has LICENSE-MIT and LICENSE-APACHE, and the README explains code/model licenses, but a short badge/summary near the top would help.
- Add a compact table of contents, since the README covers architecture, training, inference, models, licensing, and citation.
- Add a short quick-start block near the top that points users to the recommended path depending on their backend: PyTorch, MLX, MLX-Swift, or Rust.
- Add a small contributor note linking to CONTRIBUTING.md and clarifying that the project is research-oriented and mainly welcomes bug fixes.
- Minor polish: there are a couple of small typos in the README/CONTRIBUTING.md, such as "inferencde", "alignemnt", and "MHibiki".
This is only a documentation polish suggestion, not a criticism of the project. I would be happy to help with a small README cleanup PR if you are interested.
Hi Kyutai team, thanks for releasing Hibiki. I read through the README and CONTRIBUTING.md, and the project is already clearly connected to the paper, samples, Hugging Face models, Colab, and the different inference backends.
I think the README could still benefit from a small GitHub/open-source presentation pass to make the project easier to scan for new visitors. A few possible improvements:
This is only a documentation polish suggestion, not a criticism of the project. I would be happy to help with a small README cleanup PR if you are interested.