Dockerfile

# Serve an already-GGUF model with prebuilt llama.cpp binaries (no compile, no convert -> fast).
FROM python:3.11-slim

ENV DEBIAN_FRONTEND=noninteractive \
    PYTHONUNBUFFERED=1 \
    HF_HOME=/home/user/.cache/huggingface \
    HF_HUB_ENABLE_HF_TRANSFER=1

RUN apt-get update && apt-get install -y --no-install-recommends \
        curl ca-certificates libgomp1 \
    && rm -rf /var/lib/apt/lists/* \
    && useradd -m -u 1000 user

# ---- Prebuilt llama.cpp CPU binaries: llama-server + shared libs (NO build) ----
RUN mkdir -p /opt/llamabin && cd /opt/llamabin \
    && curl -fsSL -o l.tar.gz \
       https://github.com/ggml-org/llama.cpp/releases/download/b9670/llama-b9670-bin-ubuntu-x64.tar.gz \
    && tar xzf l.tar.gz --strip-components=1 \
    && rm l.tar.gz \
    && chmod +x /opt/llamabin/llama-server
ENV LD_LIBRARY_PATH=/opt/llamabin \
    PATH=/opt/llamabin:$PATH

# ---- Python deps: downloader + proxy only (no torch/transformers needed) ----
RUN pip install --no-cache-dir \
        huggingface_hub hf_transfer \
        "fastapi>=0.110" "uvicorn[standard]>=0.27" requests

WORKDIR /home/user/app
COPY app.py .
COPY chat_template.jinja .
RUN chown -R user:user /home/user

USER user
EXPOSE 7860
CMD ["python", "app.py"]