Dockerfile

# Dockerfile.hf — HuggingFace Spaces optimised image
#
# Target environment:
#   2 vCPU  |  12 GB RAM  |  16 GB disk  |  No GPU
#   Python 3.12.12  |  PyTorch CPU-only
#
# Local test:
#   docker build -f Dockerfile.hf -t healthexpert-hf .
#   docker run -p 7860:7860 --memory="12g" --cpus="2" healthexpert-hf
#
# Push to HuggingFace:
#   Build is triggered automatically when this Dockerfile is in the Space repo root
#   (rename to Dockerfile before pushing to HF).

FROM python:3.12.12-slim

# ── System dependencies ────────────────────────────────────────────────────────
# Minimal set: OCR engine + OpenCV headless libs only.
# build-essential and cmake REMOVED — llama-cpp-python installs from a pre-built
# wheel (llama-cpp-python==0.3.30, py3-none-manylinux2014_x86_64) so no C++
# compilation occurs at image build time.
RUN apt-get update && apt-get install -y --no-install-recommends \
    tesseract-ocr \
    libgl1 \
    libglib2.0-0 \
    libgomp1 \
    && rm -rf /var/lib/apt/lists/*

WORKDIR /app

# ── Python environment ─────────────────────────────────────────────────────────
# CRITICAL: Install CPU-only PyTorch FIRST to prevent pip pulling the 2.5 GB CUDA build.
# CPU wheel is ~260 MB vs 2.5 GB for CUDA — essential for 16 GB disk constraint.
RUN pip install --no-cache-dir \
    torch==2.5.1+cpu \
    torchvision==0.20.1+cpu \
    torchaudio==2.5.1+cpu \
    --index-url https://download.pytorch.org/whl/cpu

# ── Application dependencies ───────────────────────────────────────────────────
COPY requirements.txt .
RUN pip install --no-cache-dir -r requirements.txt

# ── Application codebase ───────────────────────────────────────────────────────
COPY . .

# Remove GPU-mode files to keep image clean (optional, saves ~1 MB)
RUN rm -f requirements_gpu.txt Dockerfile_bak

RUN chmod +x start.sh

# ── Environment configuration ─────────────────────────────────────────────────
# HF_MODE=1 activates low-resource path in config.py, gen_llm.py, embed_llm.py
ENV HF_MODE=1
ENV ADMIN_MODE=1
ENV PORT=7860

# HuggingFace model cache — use /app/models to keep within Space storage
ENV HF_HOME=/app/models
ENV TRANSFORMERS_CACHE=/app/models
ENV SENTENCE_TRANSFORMERS_HOME=/app/models

# Suppress noisy PyTorch/tokenizer warnings in logs
ENV PYTHONWARNINGS=ignore
ENV TOKENIZERS_PARALLELISM=false

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

# ── Entrypoint ────────────────────────────────────────────────────────────────
# -hf activates low-resource mode; ADMIN_MODE env var controls admin controls.
# To disable admin controls for public endpoint, set ENV ADMIN_MODE=0 above
# or pass -noadmin here.
ENTRYPOINT ["bash", "start.sh", "-hf"]