Skip to content

Latest commit

 

History

History
30 lines (21 loc) · 872 Bytes

README.md

File metadata and controls

30 lines (21 loc) · 872 Bytes

Idefics2-OCR

Fine-tuned the HuggingFaceM4/idefics2-8b model on the nielsr/docvqa_1200_examples_donut dataset for document VQA pairs. Checkout Idefics2-OCR on Hugging Face.

Find the rest of training details here.

Image Description

Finetune

Set your wanb token in the .env file as WANDB_API.

  • Install the requirements
pip install -r requirements.txt
  • Finetune Idefics
python3 idefics2.py --wandb True

Run the app

You don't have to finetune for running the app the model is loaded from Hugging Face.

python3 app.py