Dockerfile

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

WORKDIR /app

# Python仮想環境を作るための最低限のものを入れる
RUN apt-get update && apt-get install -y --no-install-recommends \
    python3-pip python3-venv \
    && rm -rf /var/lib/apt/lists/*

# venv を作成
RUN python3 -m venv /opt/venv
ENV PATH="/opt/venv/bin:$PATH"

# venv にだけ pip パッケージを入れる
RUN pip install --no-cache-dir -U huggingface_hub

# モデルとmmprojをダウンロード
RUN python3 -c 'from huggingface_hub import hf_hub_download; \
    repo="usermma/Holo-3.1-0.8B-abliterated-rdo-Q4_K_M-GGUF"; \
    hf_hub_download(repo_id=repo, filename="holo-3.1-0.8b-abliterated-rdo-q4_k_m.gguf", local_dir="/app")'

CMD ["--server", \
     "-m", "holo-3.1-0.8b-abliterated-rdo-q4_k_m.gguf", \
     "--host", "0.0.0.0", \
     "--port", "7860", \
     "-t", "3", \
     "--mlock", \
     "--prio", "3", \
     "--swa-full", \
     "--no-slots", \
     "--mmap", \
     "--log-disable", \
     "--no-cont-batching", \
     "--threads-http", "1", \
     "--direct-io", \
     "--no-repack", \
     "--flash-attn", "off", \
     "-c", "64000", \
     "-n", "38912"]