As a Swift developer pretty much from day one, I’m wildly excited to see the SwiftForTensorflow project moving forward. Are there any recent updates on the deployment situation? I’m my experience, that’s still the biggest pain-point for ML in production—i.e., can we get our models to mlmodel (or to onnx, then mlmodel)? And if not, is there some other way we can deploy them on a device. I have to admit that I’m not at all clear on what’s required to get a trained model running on iOS/macOS/Metal, but it would be incredibly fantastic if we could just develop and train our models in Swift, save to mlmodel, and deploy. I’m assuming a streamlined workflow like that is in the roadmap, but I’m curious where things are at now, and how for off that might be.
Exciting work, All!