Finetune help
#1
by AaryaanSing - opened
Can you share code on how you tuned it
Please share your code. I get '!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!' output after fine-tuning :(
Load the model in only bfloat16 and do not quantize it when finetunning it.
I have posted a guide on Medium:
(https://medium.com/@adhams.ka7/fine-tuning-granite-docling-a-practical-guide-to-boosting-document-ocr-accuracy-4eb6129bb59d)
Hope this helps !