Clarification on Gemma ToU applicability for models trained on this dataset
Hi, thanks for releasing this dataset!
I have a question regarding the licensing implications when using this dataset to train a new model. Since this dataset contains synthetic data generated by Gemma, I am concerned about the "Model Derivatives" definition in the Gemma Terms of Use (Section 1.1 (e)).
The Gemma ToU states that "Model Derivatives" include models created using methods based on the generation of synthetic data Outputs by Gemma. This implies that any model trained on this dataset might legally be considered a Gemma derivative, and thus bound by the Gemma ToU (and its restrictions), despite the dataset itself being licensed under ODC-BY.
Could you clarify if training a model on this finetranslations dataset triggers the Gemma ToU restrictions for the downstream model? Or is the intention of the ODC-BY license here to imply that the "output ownership" clause (Section 3.3) takes precedence, freeing users from the derivative constraints?
I would appreciate your insights on this.