[Feature] Expporting Donut to onnx and quantization

[WaterKnight1998]

Hi,

This model is very powerfull, but using it in production is hard. It would be cool if we can export it to onnxruntime and do quantization so it can be used in CPU or mobile devices.

I would be glad to help with the feature as it is Hacktoberfest :) @gwkrsrch @napatswift @SamSamhuns

Best regards, WaterKnight

[Prasanna-iq]

Any update on this @WaterKnight1998 ?

[jackkwok]

I am interested as well. Has anyone tried torch.onnx.export?

[Mhhhaster]

How about use FasterTransformer to do quantization on this model

[ambroiseb]

Hi, I managed to convert a Donut model to onnx. My work is available here. Hope it helps :)