ToneX uses Tensorflow, but I get what you mean.
There aren't any standards with this stuff. You've got weights, you've got biases. But each ML framework uses a slightly different approach. Because of commerical realities, it isn't really feasible to expect all companies to pull together and agree on a standard.
At the moment in the HISE community, there is a push to support NAM models. But NAM relies on a different ML framework than HISE does. So they're trying to bridge the gap. Once it is done, you're going to see a butt ton of NAM based plugins released by companies like Korneff and the like.
The conversation is here if you're interested:
Hi mates. I converted this simple neural network example from this page (https://medium.com/technology-invention-and-more/how-to-build-a-simple-neural-networ...
forum.hise.audio