First of all thanks to Explosion.ai for the amazing tools you provide.
By looking at the MIT license, it is clear that spaCy can be reused within a proprietary software. But the intellectual property mechanism around a pre-trained spaCy model that has been fine-tuned using Prodigy is less obvious to me.
Does a company still own the IP of a model trained or fine-tuned with annotations that have been obtained via Prodigy?
Hi @qthurier,
If I understand you correctly, using prodigy to fine-tune an existing model wouldn’t change that model’s license in any way. As long as you own prodigy you can use it to work with whatever models you want and use their existing licenses.
Yes, you’re only using Prodigy for creating the labelled data, so the labelled data will still be yours (assuming your company also own the IP of the raw data). When you fine-tune a model with Prodigy, it still just calls into nlp.update under the hood. Using a commercial tool to help you update your model more effectively doesn’t change the model license in any way.
(You can also think about it this way: If you use Excel to calculate something using proprietary data, Microsoft doesn’t own the resulting spreadsheet. If you cut and edit a video using proprietary recordings, the resulting file is still 100% yours, even if you used a commercial video editing tool.)