#501 License compatibility when serving TimesFM 3.0 architecture via third-party libraries (e.g. Darts)

Hi! Thanks for the great TimesFM 3.0 release.

The Darts forecasting library (unit8co/darts) is considering adding support for TimesFM 3.0 (issue unit8co/darts#3196). We noticed that the repo source code is Apache-2.0, but the TimesFM 3.0 checkpoints on Hugging Face (google/timesfm-3.0-pytorch) are distributed under the timesfm-non-commercial-license-v1.0.

A couple of questions to clarify before integration:

If a library like Darts only wraps/serves the architecture (code, Apache-2.0) and the user still downloads the weights directly from Hugging Face, would that library and its non-commercial users be considered compliant? i.e. does the non-commercial restriction apply only to the checkpoint itself?
Are there plans to release TimesFM 3.0 weights under a license that allows commercial use (like the Apache-2.0 weights of earlier versions), or to offer an opt-in commercial variant?
Would you be open to documenting this more explicitly (e.g. in the model card) so that integration libraries can surface the right license notice to users?

For context, Darts already handles similar cases with an explicit accept_license flag (default False) that raises if the user hasn't accepted the license (see the TiRex integration). We'd like to follow the same approach if it's the right path here.