This repository includes the following:
- NER - extract cyber entities. Evaluation of different models
- Image-Captioning - caption desktop, screenshots and regular images. Evaluation of different models
- Playground - Export the selected NER and Image-Captioning models via streamlit
- Triton - Incomplete efforts to improve models inference times using Nvidia Triton
- In
NER
folder:
./build/export.sh
- In
Image-Captioning
folder:
./build/download_lavis_model.sh
- In
Image-Captioning/build/
folder:
git clone [email protected]:huggingface/optimum.git
- In
Image-Captioning/build/optimum
folder:
python3 setup.py develop
cp /usr/local/bin/optimum-cli .
- In
Image-Captioning/build
folder:
./optimum/optimum-cli export onnx -m google/pix2struct-screen2words-base --optimize O3 onnx/pix2struct_base
- In root folder:
docker-compose build && docker-compose up
- Access the playground via http://0.0.0.0:8501/