I have very little knowledge about the actual license law, but what is the right way to do that?
Let's say there is an open source project published with a very permissive license that has one function you need. Putting the whole project as a dependency is an overkill and rewriting the function with your own variable names and indentation doesn't sound like an improvement.
Would adding a copyright notice somewhere in the repository be enough?
How does tie in with CoreML that was announced today, which also uses Metal under the hood and supports models built with Tensorflow/Keras and the likes.
As it uses Metal under the hood, it takes advantage of the device GPU, so it runs fast and supports real-time applications.
It also allows to run TensorFlow models, so it should be easy to run your existing model.
If you guys could provide feedback, it would be cool!