Dockerfile

FROM ghcr.io/livebook-dev/livebook:latest-cuda12.1

ENV LIVEBOOK_APP_SERVICE_NAME="🐳 Hugging Face - $SPACE_TITLE"
ENV LIVEBOOK_APP_SERVICE_URL="https://huggingface.co/spaces/$SPACE_AUTHOR_NAME/$SPACE_REPO_NAME"
ENV LIVEBOOK_UPDATE_INSTRUCTIONS_URL="https://livebook.dev"
ENV LIVEBOOK_WITHIN_IFRAME="true"
ENV LIVEBOOK_APPS_PATH="/public-apps"
ENV LIVEBOOK_APPS_PATH_WARMUP="manual"
ENV LIVEBOOK_DATA_PATH="/data"
ENV LIVEBOOK_PORT="7860"

EXPOSE 7860

RUN mkdir -p /data
RUN chmod 777 /data

# The Space container runs with user ID 1000
RUN useradd -m -u 1000 user
ENV HOME="/home/user"

USER user

# TODO: remove this once EXLA v0.8 is out and we infer the default from the environment
# Configure Nx/EXLA to use the GPU (also relevant for warmup_apps below)
ENV XLA_TARGET="cuda120"

COPY --chown=user public-apps/ /public-apps
RUN /app/bin/warmup_apps