Move executorch export to optimum-executorch

[echarlaix]

Following huggingface/optimum#2090 (comment)

ExecuTorch export (introduced in huggingface/optimum#2090) is moved in optimum-executorch see huggingface/optimum#2165

[michaelbenayoun]

LGTM!
Thanks for handling this

[guangy10]

@echarlaix do you mind explaining the motivation of moving to optimum-executorch? I remember you briefly mentioned that during out meetup at HF Paris office but can't recall the details. I want to make sure such a move will not potentially decrease the visibility executorch to HF users. Thanks!

[echarlaix]

@guangy10 we are planning to keep in the optimum repo only things which are common to all subpackages and to move everything else in specific repos (the same will happen for anything onnx related for example). I understand your point concerning visibility, something we can do to make sure this is not the case is to clearly redirect users to this repo in the optimum and transformers documentation, what do you think ?

[guangy10]

@guangy10 we are planning to keep in the optimum repo only things which are common to all subpackages and to move everything else in specific repos (the same will happen for anything onnx related for example). I understand your point concerning visibility, something we can do to make sure this is not the case is to clearly redirect users to this repo in the optimum and transformers documentation, what do you think ?

Thanks for the clarification. With such move, is it possible for us to get more fine-grained metrics/stats? For example, compare DAU/downloads/checkout across different optimum supported backends, e.g. onnx, executorch, tflite, etc. assuming onnx and tflite are going to optimum-onnx, and optimum-tflite.
Can we also track per model inference across different backends? Maybe other metrics, do you know?