Dockerfile.cuda

# NOTE: This Docker file is configured to deploy Type4Py on our server and GPUs.
# For us, these configs seem to work: CUDA 11.0.3, ONNX v1.10.0, nvidia driver 450.36.06
FROM nvidia/cuda:11.0.3-cudnn8-runtime-ubuntu20.04
WORKDIR /type4py/
# Put the required models files in a folder "t4py_model_files" inside "/type4py"
# -type4py/
# --type4py/
# --t4py_model_files/
COPY . /type4py
ENV T4PY_LOCAL_MODE="1"

RUN apt update --fix-missing -y && apt upgrade -y && apt install -y python3-pip libpq-dev

# The current model files are pickled with the below ver. of sklearn
RUN pip install scikit-learn==0.24.1

# Install Annoy from a pre-built binary wheel to avoid weird SIGILL error on some systems
RUN pip install https://type4py.com/pretrained_models/annoy-wheels/annoy-1.17.0-cp38-cp38-linux_x86_64.whl

# For production env., install ONNXRuntime with GPU support
RUN pip install onnx==1.10 onnxruntime==1.10 onnxruntime-gpu==1.10

# Install Type4Py
RUN pip install -e .

# Web server's required packages
RUN pip install -r type4py/server/requirements.txt

# Install NLTK corpus
RUN python3 -c "import nltk; nltk.download('stopwords')"
RUN python3 -c "import nltk; nltk.download('wordnet')"
RUN python3 -c "import nltk; nltk.download('omw-1.4')"
RUN python3 -c "import nltk; nltk.download('averaged_perceptron_tagger')"

WORKDIR /type4py/type4py/server/

EXPOSE 5010

CMD ["bash", "run_server.sh"]