Dockerfile

# Moonley schema-v5 backend — Hugging Face Space (Docker SDK), CPU-only.
FROM python:3.11-slim

ENV PYTHONUNBUFFERED=1 \
    PIP_NO_CACHE_DIR=1 \
    HF_HOME=/tmp/hf \
    SENTENCE_TRANSFORMERS_HOME=/tmp/hf/sentence-transformers \
    TRANSFORMERS_CACHE=/tmp/hf/transformers \
    MOONLEY_DATA=/tmp/moonley_release \
    MOONLEY_STATUTE="/app/statute corpus" \
    MOONLEY_STATUTE_CHROMA=/tmp/moonley_statutes \
    MOONLEY_QWEN_MODEL=/app/qwen_model \
    MOONLEY_QWEN_DTYPE=bfloat16 \
    MOONLEY_WARM_QUERY_MODEL=1 \
    MOONLEY_DEEP=never \
    MOONLEY_SKIM=0 \
    MOONLEY_HELD_ARM=0 \
    MOONLEY_CITECTX=0 \
    MOONLEY_BUDGET_S=20 \
    MOONLEY_DEVICE=cpu \
    MOONLEY_LOG_DIR=/tmp/moonley_logs \
    MOONLEY_PDF_CACHE=/tmp/pdf_cache \
    MOONLEY_KEYWORD=0 \
    OMP_NUM_THREADS=8 \
    TOKENIZERS_PARALLELISM=false

WORKDIR /app

RUN apt-get update && apt-get install -y --no-install-recommends build-essential fonts-dejavu-core tesseract-ocr \
    && rm -rf /var/lib/apt/lists/*

# CPU torch in its own layer (heavy, rarely changes), then the serving deps.
RUN pip install torch --index-url https://download.pytorch.org/whl/cpu
COPY phase1/deploy/requirements.txt ./req.txt
RUN pip install -r req.txt huggingface_hub

# Fetch private artifacts at runtime so neither snapshot is embedded in this
# public image. Revisions are immutable and every expected entrypoint is checked.
ENV MOONLEY_RELEASE_REPO=vg15o2/themis-indian-kanoon-qwen-v1 \
    MOONLEY_RELEASE_REVISION=12f58201987cc8ec7697010754ab75765c5f5a24 \
    MOONLEY_STATUTE_REPO=vg15o2/themis-statutes-v1 \
    MOONLEY_STATUTE_REVISION=ebf66528e417358a09903d95f6718ccfeb94a426

ARG QWEN_MODEL_REVISION=5cf2132abc99cad020ac570b19d031efec650f2b
RUN python -c "import os; from huggingface_hub import snapshot_download; snapshot_download(repo_id='Qwen/Qwen3-Embedding-4B', repo_type='model', revision=os.environ['QWEN_MODEL_REVISION'], local_dir='/app/qwen_model')"

# Keep private release pointers after the large, cacheable model layer. Updating
# a small template release must not force the Qwen model to download again.
ENV MOONLEY_DRAFTING_TEMPLATE_REPO=vg15o2/themis-drafting-templates-v1 \
    MOONLEY_DRAFTING_TEMPLATE_REVISION=2b036d4bef7ebe7a3b3bb8d0a094d7fefdd8c4d2

COPY . .

# HF runs the container as a non-root user; only /tmp is writable.
RUN mkdir -p /tmp/hf /tmp/moonley_logs /tmp/pdf_cache && chmod -R 777 /tmp/hf /tmp/moonley_logs /tmp/pdf_cache

EXPOSE 7860
CMD ["python", "phase1/scripts/start_private_space.py"]