This demo showcase the use of onnxruntime-rs with a GPU on CUDA 11 to run Bert in a data pipeline with Rust.

• Linux x86_64
• NVIDIA GPU with CUDA 11 (Not sure if CUDA 10 works)
• Rust (obviously)
• git lfs. Install it with https://git-lfs.github.com/

Download the data https://www.kaggle.com/c/predict-closed-questions-on-stack-overflow/data?select=train_October_9_2012.csv -> train_October_9_2012.csv and put it in the folder training.

cd training
pip install -r requirement.txt

python pytorch-model.py

I have not done any parameter search to find an optimal model as it is not the point of this repo.

I have limited the training to 500 rows to iterate faster.

cd src
python python_alternative.py

export ORT_USE_CUDA=1
cargo build --release

cargo run --release

or after build:

export LD_LIBRARY_PATH=path/to/onnxruntime-linux-x64-gpu-1.8.0/lib:${LD_LIBRARY_PATH}
./target/release/machine-learning