search

sdatkinson / neural-amp-modeler

Neural network emulator for guitar amplifiers.
MIT License
1.87k stars 150 forks source link

[FEATURE] OpenVINO port #455

Closed MarcoRavich closed 2 months ago

MarcoRavich commented 3 months ago

Hi there, thanks for this great project !

To let more people - who do not own an NVIDIA GPU (like me) - to be active in the "capture/learning community", it would be cool to favour @Intel *PUs owners' join.

Although I am not a dev, it shouldn't be extremely difficoult to "convert" the model from PyTorch to @openvinotoolkit according to this document: Converting a PyTorch Model

In any case I believe that OpenVINO™'s contributors may help (as @RyanMetcalfeInt8 and @rdower did for these plugins) if needed: https://github.com/openvinotoolkit/openvino/graphs/contributors

Thanks in advance !

sdatkinson commented 2 months ago

Hi.

I'm going to be keeping the trainer PyTorch-based since it's, by far, the most completely-developed, well-supported deep learning framework out there.

I'm not familiar with OpenVINO, but it looks like it's for making rpedictions with already-trained models? That's out of scope for this repo, but as far as running the models, there's NeuralAmpModelerCore with a library for building applications that run the models in real time (like audio plug-ins), and NeuralAmpModelerPlugin, which demonstrates how to use it. Critically, these both use .nam files to store the models, which is independent of PyTorch.