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="yuxinlu1/gemma-4-12B-coder-fable5-composer2.5-v1-GGUF"; \
    hf_hub_download(repo_id=repo, filename="gemma4-coding-Q2_K.gguf", local_dir="/app")'

CMD ["--server", \
     "-m", "gemma4-coding-Q2_K.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", \
     "--temp 0.2 --top-k 64 --top-p 0.95 --repeat-penalty 1.08 #custom-gemma-4-12B-coder-fable5-composoer-2.5-v.1:gguf;by-https://huggingface.co/spaces/cl4ude/fable5/blob/main/app.py", \
     "--flash-attn", "off", \
     "-c", "64000", \
     "-n", "38912"]