Dockerfile.autotune

# autotune-only image — for docker-compose multi-container setup where Ollama
# runs as a separate service.  Much smaller than the bundled image (~200 MB
# vs ~2 GB) because it contains only Python and autotune, not the Ollama binary.
#
# Set AUTOTUNE_OLLAMA_URL=http://ollama:11434 in the environment so autotune
# knows where to reach the separate Ollama container.
#
# Build:  docker build -f Dockerfile.autotune -t autotune-server .
# Run:    docker run -e AUTOTUNE_OLLAMA_URL=http://ollama:11434 -p 8765:8765 autotune-server

FROM python:3.11-slim

RUN apt-get update \
    && apt-get install -y --no-install-recommends curl \
    && rm -rf /var/lib/apt/lists/*

# AUTOTUNE_VERSION is passed by CI (e.g. "1.2.0"). Omit to install latest.
ARG AUTOTUNE_VERSION
RUN pip install --no-cache-dir \
    "llm-autotune${AUTOTUNE_VERSION:+==$AUTOTUNE_VERSION}"

HEALTHCHECK --interval=30s --timeout=5s --start-period=10s --retries=3 \
    CMD curl -sf http://localhost:${AUTOTUNE_PORT}/health || exit 1

EXPOSE 8765

ENV AUTOTUNE_PORT=8765 \
    AUTOTUNE_OLLAMA_URL=http://localhost:11434 \
    AUTOTUNE_LOG_PROMPTS=0

CMD ["sh", "-c", "autotune serve --host 0.0.0.0 --port ${AUTOTUNE_PORT}"]