Enable KittenTTS on ExecuTorch

[mergennachin]

Hi, I'm one of the maintainers of ExecuTorch repo -- https://github.com/pytorch/executorch

I see that you have enabled a great edge model.

Any plans to support torch.export and ExecuTorch-compatible model?

I see that you already have an Onnx version, which is great, but was wondering if you want to expand this.

Other prominent libraries such as Yolo are adding, for example:

https://github.com/ultralytics/ultralytics/pull/22244/files#diff-2e48775a7706bdc3f9f597330f608840efd012a0396a18968aa8d9f3c266f819

[olokobayusuf]

@mergennachin I reached out the founders here and haven't heard back (we recently added support for the model on Muna, using ORT). It did give me one idea I'm sure you all have considered: ONNX -> PTE. Any thoughts on this? It would instantly widen the number of models that ET can support on day one.

For example, Muna's support for QNN relies on exporting from PyTorch to ONNX, then using QAIRT to convert it to Qualcomm's downstream format.