Is it possible to fine tune with our own datasets?

[ninedesu]

I want to know if we can use our own dataset to finetune the OCR

[PeterStaar-IBM]

@ninedesu This is an excellent question, and yes, we plan to build a community where people can contribute data for fine-tuning. At the moment, we are gathering all our internal and external datasets (eg https://huggingface.co/datasets/ds4sd/DocLayNet) and preparing them so we can share them all on the huggingface website!

With regard to OCR, we have a bit of work to do and are right now relying on 3rd party OCR.

[bit-scientist]

@PeterStaar-IBM, is there any update on custom training guidelines?

[MengFoong]

@PeterStaar-IBM would glad to hear an updates

[PeterStaar-IBM]

we just released the first test dataset (https://huggingface.co/datasets/ds4sd/docling-dpbench). More will come soon.

[FrankFacundo]

@PeterStaar-IBM I made some scripts to fine-tune the layout model, in which directory could I add them if possible ?