Dockerfile

FROM python:3.11-slim

LABEL maintainer="local-llm-api"
LABEL description="Local LLM inference API — llama-cpp-python + Flask (multi-model)"

WORKDIR /app

# ── System deps ───────────────────────────────────────────────────────────────
RUN apt-get update && apt-get install -y --no-install-recommends \
        curl \
        ca-certificates \
        build-essential \
    && rm -rf /var/lib/apt/lists/*

# ── Copy project files ────────────────────────────────────────────────────────
COPY . .
RUN chmod +x install.sh start.sh

# ── Environment ───────────────────────────────────────────────────────────────
# PORT=7860 for HuggingFace Spaces. Override at runtime for other hosts.
# Set ACTIVE_MODELS to control which models load (comma-separated names).
# Example:  docker run -e ACTIVE_MODELS="Qwen3.5-4B,DeepSeek-R1-8B" ...
ENV PORT=7860
ENV PYTHONUNBUFFERED=1

# ── Python deps ───────────────────────────────────────────────────────────────
# Install llama-cpp-python from CPU wheel index first; fall back to PyPI build.
RUN pip install --no-cache-dir \
        flask \
        huggingface-hub \
        numpy \
        chromadb \
        sentence-transformers \
    && ( pip install --no-cache-dir llama-cpp-python \
            --extra-index-url https://abetlen.github.io/llama-cpp-python/whl/cpu \
         || pip install --no-cache-dir llama-cpp-python )

# ── Override port for HuggingFace Spaces (7860 instead of 5000) ──────────────
#RUN python3 -c "
#import json
#with open('install.json') as f: c = json.load(f)
#c.setdefault('server', {})['port'] = 7860
#with open('install.json', 'w') as f: json.dump(c, f, indent=2)
#"

# ── Model volume ──────────────────────────────────────────────────────────────
# Mount models here at runtime:
#   docker run -v /host/models:/app/model ...
# Each model lives in /app/model/<ModelName>/chunks/*.gguf
VOLUME ["/app/model"]

# ── Port ─────────────────────────────────────────────────────────────────────
EXPOSE 7860

# ── Health check ──────────────────────────────────────────────────────────────
# Generous start-period — models can take 1-5 min to join+load from disk.
HEALTHCHECK --interval=30s --timeout=15s --start-period=600s --retries=10 \
    CMD curl -sf "http://localhost:${PORT}/health" | python3 -c \
        "import sys,json; d=json.load(sys.stdin); sys.exit(0 if d.get('status') in ('ok','partial') else 1)" \
    || exit 1

# ── Entrypoint ────────────────────────────────────────────────────────────────
# start.sh: installs any missing packages, syncs/downloads models, starts Flask.
CMD ["bash", "start.sh"]