Dockerfile

# Debian slim base, glibc compatible
FROM python:3.12-slim

# HF cache 指向可写目录
ENV HF_HOME=/tmp/.cache
ENV XDG_CACHE_HOME=/tmp/.cache

# 安装系统依赖 - 使用 libopenblas-dev 而不是 libblas3
RUN apt-get update && apt-get install -y --no-install-recommends \
        ca-certificates \
        git \
        wget \
        libgomp1 \
        libstdc++6 \
        curl \
        libopenblas-dev \   
        liblapack-dev \      
    && rm -rf /var/lib/apt/lists/*

# 升级 pip / setuptools / wheel
RUN pip install --no-cache-dir --upgrade pip setuptools wheel

# 复制并安装本地 CPU-only wheel（linux_x86_64 + Python 3.12）
COPY llama_cpp_python-0.3.16-cp312-cp312-linux_x86_64.whl /tmp/
RUN pip install --no-cache-dir /tmp/llama_cpp_python-0.3.16-cp312-cp312-linux_x86_64.whl \
    && rm -f /tmp/llama_cpp_python-0.3.16-cp312-cp312-linux_x86_64.whl

# 安装 Python 依赖
RUN pip install --no-cache-dir gradio huggingface-hub

# 环境变量
ENV MODEL_DIR="/models"
ENV MODEL_FILE="Qwen3_Medical_GRPO.i1-Q4_K_M.gguf"
ENV MODEL_REPO="mradermacher/Qwen3_Medical_GRPO-i1-GGUF"

# 创建可写目录
RUN mkdir -p ${MODEL_DIR} /tmp/.cache \
    && chmod -R 0777 ${MODEL_DIR} /tmp/.cache

# 复制 app.py
COPY app.py /app/app.py
WORKDIR /app

# 暴露端口
EXPOSE 7860

# 启动
CMD ["python", "app.py"]