Commercial redistribution and training-data license clarification for Glyph
#1
by Aghlara - opened
Hello,
We are evaluating ThingAI/Glyph for inclusion in a commercial software platform.
We understand the Glyph model repository is published under Apache 2.0, and we would like to confirm the intended licensing of the published model weights and converted forms.
Could you please confirm that:
- The published Glyph model weights may be used commercially and redistributed under Apache 2.0.
- We may convert the model to ONNX and redistribute that converted model as part of our proprietary commercial software product, while complying with Apache 2.0 attribution and notice requirements.
- None of the datasets used to train Glyph, including textdetox/multilingual_toxicity_dataset, impose additional downstream use, redistribution, copyleft, or source-disclosure restrictions on the published Glyph model weights.
- There are no additional model-specific licensing obligations beyond those stated under Apache 2.0 that would affect commercial redistribution.
We would not redistribute the training datasets themselves, only the pretrained/converted model artifacts.
A written confirmation would help us complete our internal release review.
Thank you,
Aghlara changed discussion status to closed
Hi @Aghlara ,
Thanks for your interest in Glyph! Happy to confirm:
- Yes — the published Glyph model weights are released under Apache 2.0. You may use and redistribute them commercially.
- Yes — you may convert the model to ONNX (or any other format) and redistribute the converted artifacts as part of your commercial product, as long as you comply with Apache 2.0 attribution and notice requirements.
- The training datasets (including textdetox/multilingual_toxicity_dataset) do not impose additional downstream restrictions on the published model weights.
- Correct — there are no additional model-specific licensing obligations beyond Apache 2.0.
Feel free to reopen this discussion or reach out if you need anything else.
Best,
Michelangelo