Dockerfile

FROM nvidia/cuda:11.8.0-devel-ubuntu22.04
RUN apt update && apt install -y \
    git \
    build-essential \
    libopenblas-dev \
    wget \
    python3-pip \
    nodejs \
    npm 

RUN useradd -m -u 1000 user

USER user

ENV HOME=/home/user \
    PATH=/home/user/.local/bin:$PATH


WORKDIR $HOME/app
COPY --chown=user . $HOME/app

RUN python3 -m pip install aphrodite-engine
RUN pip install huggingface-hub hf-transfer

ENV HF_HUB_ENABLE_HF_TRANSFER=1
RUN huggingface-cli download TheBloke/goliath-120b-GPTQ --local-dir $HOME/goliath-gptq --local-dir-use-symlinks False --cache-dir $HOME/cache


EXPOSE 7860
# Should expose on all network interfaces so 0.0.0.0 not only localhost
CMD /bin/bash -c "/bin/python3 -m aphrodite.endpoints.openai.api_server $ENGINE_ARGS --port 7860 --host 0.0.0.0 --model ~/goliath-gptq"