Dockerfile

FROM ghcr.io/ggml-org/llama.cpp:full

WORKDIR /app

RUN apt update && apt install -y python3 python3-pip python3-venv curl
RUN python3 -m venv /opt/venv
ENV PATH="/opt/venv/bin:$PATH"

RUN pip install -U pip huggingface_hub
RUN pip install -U fastapi "uvicorn[standard]" httpx python-multipart pillow

ARG HF_TOKEN=""
ENV HF_TOKEN=$HF_TOKEN

RUN python3 -c 'from huggingface_hub import hf_hub_download; \
    repo="unsloth/gemma-4-E2B-it-GGUF"; \
    hf_hub_download(repo_id=repo, filename="gemma-4-E2B-it-UD-Q5_K_XL.gguf", local_dir="/app"); \
    hf_hub_download(repo_id=repo, filename="mmproj-BF16.gguf", local_dir="/app")'

ENV HF_TOKEN=""

COPY main.py start.sh /app/
RUN chmod +x /app/start.sh

ENTRYPOINT []
CMD ["/app/start.sh"]