FROM ghcr.io/ggml-org/llama.cpp:server-cuda

ARG UID=1000
ARG GID=1000
RUN groupadd --system --gid ${GID} worker && \
    useradd --system --gid ${GID} --uid ${UID} --home /home/worker worker && \
    mkdir -p /home/worker && chown worker:worker /home/worker

WORKDIR /home/worker
USER worker
ENTRYPOINT ["/app/llama-server"]