Dockerfile

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

WORKDIR /app

# パッケージのインストール
RUN apt update && apt install -y python3-pip
RUN pip install -U huggingface_hub

# モデルのダウンロード
RUN python3 -c 'from huggingface_hub import hf_hub_download; \
    hf_hub_download(repo_id="mistralai/Ministral-3-3B-Instruct-2512-GGUF", \
    filename="Ministral-3-3B-Instruct-2512-Q4_K_M.gguf", \
    local_dir="/app", \
    local_dir_use_symlinks=False)'

# 設定
CMD ["--server", \
     "-m", "/app/Ministral-3-3B-Instruct-2512-Q4_K_M.gguf", \
     "--host", "0.0.0.0", \
     "--port", "7860", \
     "-t", "2", \
     "-c", "128000", \
     "-n", "32000"]