Dockerfile

# Vaaani Flagship — CPU-only inference Space (free HF tier: 2 vCPU, 16 GB RAM).
# Serves the Qwen2.5-3B base GGUF + the curriculum LoRA via llama-cpp-python.
# NO GPU. The CPU-only design is the moat AND what makes free hosting possible.
FROM python:3.11-slim

ENV PYTHONUNBUFFERED=1 \
    HF_HOME=/tmp/hf-cache \
    OMP_NUM_THREADS=2

WORKDIR /app

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

COPY requirements.txt .
# Prebuilt CPU wheels keep the image build fast (no local llama.cpp compile).
RUN pip install --no-cache-dir -r requirements.txt \
        --extra-index-url https://abetlen.github.io/llama-cpp-python/whl/cpu

COPY . .

# HF Spaces route traffic to port 7860.
EXPOSE 7860
CMD ["uvicorn", "app:app", "--host", "0.0.0.0", "--port", "7860"]