# Ubuntu 24.04 base to match the glibc of the official llama.cpp release binary. FROM ubuntu:24.04 ENV DEBIAN_FRONTEND=noninteractive # Runtime deps only (no compiler/cmake — prebuilt binaries). RUN apt-get update && apt-get install -y \ curl \ ca-certificates \ libgomp1 \ libcurl4 \ && rm -rf /var/lib/apt/lists/* # --------------------------------------------------------------------------- # Prebuilt llama.cpp CPU server (per-microarch variants selected at runtime). # --------------------------------------------------------------------------- ARG LLAMA_TAG=b9895 RUN mkdir -p /llama.cpp/build/bin \ && curl -fsSL -o /tmp/llama.tar.gz \ "https://github.com/ggml-org/llama.cpp/releases/download/${LLAMA_TAG}/llama-${LLAMA_TAG}-bin-ubuntu-x64.tar.gz" \ && tar -xzf /tmp/llama.tar.gz -C /llama.cpp/build/bin --strip-components=1 \ && rm /tmp/llama.tar.gz \ && chmod +x /llama.cpp/build/bin/llama-server \ && /llama.cpp/build/bin/llama-server --version 2>&1 | head -5 ENV LD_LIBRARY_PATH=/llama.cpp/build/bin # --------------------------------------------------------------------------- # llama-swap: model-swapping proxy — routes by the request's "model" field. # --------------------------------------------------------------------------- ARG SWAP_VERSION=236 RUN curl -fsSL -o /tmp/llama-swap.tar.gz \ "https://github.com/mostlygeek/llama-swap/releases/download/v${SWAP_VERSION}/llama-swap_${SWAP_VERSION}_linux_amd64.tar.gz" \ && tar -xzf /tmp/llama-swap.tar.gz -C /usr/local/bin llama-swap \ && rm /tmp/llama-swap.tar.gz \ && chmod +x /usr/local/bin/llama-swap \ && /usr/local/bin/llama-swap --version WORKDIR /app # Models are pulled at container startup by entrypoint.sh (keeps builds fast). RUN mkdir -p /models && chmod -R 777 /models COPY config.yaml entrypoint.sh ./ RUN chmod +x /app/entrypoint.sh EXPOSE 7860 # Two ~1.3GB models are pulled on first start. HEALTHCHECK --interval=30s --timeout=10s --start-period=420s --retries=3 \ CMD curl -f http://localhost:7860/health || exit 1 CMD ["/app/entrypoint.sh"]