Support for microsoft/Florence-2-large-ft vision model

[AshD]

microsoft/Florence-2-large-ft works really well and it is much smaller than Phi-3 vision.

Would it be possible to convert this model in Onnx format?

Thanks,
Ash

[tarzain]

+1! Would love to have this in ONNX

[balachandarsv]

+1 It would be great to have this in ONNX.

[aciddelgado]

Thanks all for the suggestion; we'll add this model to our roadmap and we'll discuss the priority internally. Thanks!

[AshD]

Xenova created an Onnx version of the Florence model
https://huggingface.co/onnx-community/Florence-2-base-ft/tree/main/onnx

[timothycarambat]

@AshD Been looking for that, thank you!

[trincadev]

Xenova created an Onnx version of the Florence model https://huggingface.co/onnx-community/Florence-2-base-ft/tree/main/onnx

Thank you. Any news for support on python? Xenova's onnx florence-2 model requires transformers.js

[natke]

Thank you for your interest in Florence-2 V. We are looking into it. Moving this issue into Discussions

[Youho99]

+1

[callmehanyu]

+1

[qingfengcss]

Could the author write a document about how to convert florence model to onnx format?

[Youho99]

Add same support request for new phi-3.5-vision
Discussion here : #827

[happyme531]

I made an onnxruntime inference script for Florence2:
https://huggingface.co/happyme531/Florence-2-base-ft-ONNX-RKNN2