# Model license and attribution

The files `model.q4.safetensors`, `tokenizer-e351c8d8-checkpoint125.safetensors`,
`tokenizer_spm_32k_3.model`, and the files under `voices/` are derived from
`kyutai/personaplex-rl-seamless` at revision
`3fa800309a4b743a8a6d764253eb45def0334afc`.

The upstream model card describes this lineage:

| Layer | Source | License |
| --- | --- | --- |
| Moshi | `kyutai/moshiko-pytorch-bf16` | [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/) |
| PersonaPlex delta | `nvidia/personaplex-7b-v1` | [NVIDIA Open Model License](https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/) |
| RL-Seamless delta | `kyutai/personaplex-rl-seamless` | [CC BY-NC 4.0](https://creativecommons.org/licenses/by-nc/4.0/) |

The final merged and quantized weights require compliance with both CC BY-NC 4.0 and the
NVIDIA Open Model License. Review the full terms at the links above. The authoritative
upstream notice is in the
[`kyutai/personaplex-rl-seamless` model card](https://huggingface.co/kyutai/personaplex-rl-seamless/blob/main/README.md#license).

This conversion changes the numeric representation of eligible weights to affine Q4. It does
not change ownership, attribution, acceptable use, or license obligations.

The original runtime and conversion code distributed alongside the weights is available under
the MIT license in `LICENSE-CODE`. That license applies only to the original code.