I’m considering bundling a locally runnable DeepSeek-derived model with a paid application. The download I’m evaluating is a community quantization of a distilled checkpoint, not a file published directly by DeepSeek.

That leaves me with several places to look:

1. The original DeepSeek release and its license.
2. The exact distilled checkpoint, including any terms associated with its base model.
3. The community repository distributing the quantized weights.
4. The inference runtime I would package with the application.

For commercial model redistribution, which documents establish the permissions and notice requirements? A license badge on the final download page feels like too little information to rely on.

I understand that hosting a model internally and distributing its weights to customers may raise different questions. I’m specifically asking about distribution: what records should I keep, and how do I trace the model license provenance when a repository’s description is incomplete?