Dockerfile

# Hugging Face Spaces Dockerfile for vd project (优化版本)
# 基于 Python 3.10 slim 镜像
FROM python:3.10-slim

# 设置元数据
LABEL maintainer="opq007"
LABEL description="VD - 视频处理服务"
LABEL version="3.0.0"

# 设置工作目录
WORKDIR /app

# 设置环境变量
# 注意：以下密钥相关的环境变量可以在 Hugging Face Spaces 设置中配置
# 这样可以避免在代码中硬编码敏感信息
#
# 认证密钥配置（生产环境必须在 HF Spaces 设置中修改）：
# - API_TOKEN: API 认证令牌
# - SECRET_KEY: JWT 加密密钥
#
# 第三方服务密钥：
# - ZHIPU_API_KEY: 智谱 AI API 密钥
ENV PYTHONUNBUFFERED=1 \
    DEBIAN_FRONTEND=noninteractive \
    PYTHONDONTWRITEBYTECODE=1 \
    PIP_NO_CACHE_DIR=1 \
    PIP_DISABLE_PIP_VERSION_CHECK=1 \
    PORT=7860 \
    HOST=0.0.0.0 \
    ENABLE_GRADIO_UI=true \
    FW_MODEL=small \
    FW_DEVICE=cpu \
    FW_COMPUTE=int8 \
    VOX_ONNX_DEVICE=cpu \
    VOX_ONNX_OPTIMIZE=1 \
    VOX_ONNX_DTYPE=fp32 \
    LOG_LEVEL=INFO \
    MODELSCOPE_CACHE=/app/models/modelscope_cache

# 安装系统依赖（最小化安装）
RUN apt-get update && apt-get install -y --no-install-recommends \
    ffmpeg \
    aria2 \
    p7zip-full \
    git \
    curl \
    ca-certificates \
    && rm -rf /var/lib/apt/lists/* \
    && apt-get clean

# 设置 GitHub 仓库地址（可通过环境变量覆盖）
ARG REPO_URL=https://github.com/opq007/vd.git
ARG REPO_BRANCH=main

# 升级 pip 并安装 Python 依赖（分批安装以优化构建缓存）
RUN pip install --no-cache-dir --upgrade pip setuptools wheel && \
    pip install --no-cache-dir --default-timeout=100 \
    "faster-whisper" \
    "fastapi" \
    "uvicorn[standard]" \
    "python-multipart" \
    "aiofiles" \
    "PyJWT" \
    "gradio" \
    "requests" && \
    pip install --no-cache-dir --default-timeout=100 \
    "playwright>=1.40.0" \
    "opencv-python-headless" \
    "numpy" \
    "torch" \
    "Pillow>=10.0.0" \
    "soundfile" \
    "transformers>=4.30.0" \
    "onnxruntime" \
    "modelscope" && \
    pip install --no-cache-dir --default-timeout=100 \
    "sentencepiece" \
    "protobuf" \
    "accelerate" \
    "einops" \
    "omegaconf" \
    "scipy" \
    "librosa"

# 克隆项目代码（自动获取最新版本）
RUN echo "Cloning project from ${REPO_URL} (branch: ${REPO_BRANCH})..." && \
    git clone --depth 1 --branch ${REPO_BRANCH} ${REPO_URL} /tmp/vd && \
    cp -r /tmp/vd/* . && \
    cp -r /tmp/vd/.* . 2>/dev/null || true && \
    rm -rf /tmp/vd && \
    echo "Project cloned successfully"

# 创建必要的目录
RUN mkdir -p models uploads output debug logs && \
    chmod -R 755 models uploads output debug logs

# 下载模型文件（使用 aria2 加速，添加重试机制）
# 1. 下载 RMBG-1.4 模型（背景移除）
RUN echo "Downloading RMBG-1.4 model..." && \
    aria2c -x 16 -s 16 --max-tries=5 --retry-wait=5 \
    -d models/ -o rmbg-1.4.onnx \
    https://modelscope.cn/models/AI-ModelScope/RMBG-1.4/resolve/master/onnx/model.onnx || \
    (echo "RMBG-1.4 download failed, retrying..." && \
     aria2c -x 16 -s 16 --max-tries=10 --retry-wait=10 \
     -d models/ -o rmbg-1.4.onnx \
     https://modelscope.cn/models/AI-ModelScope/RMBG-1.4/resolve/master/onnx/model.onnx)

# 2. 下载 faster-whisper-small 模型
RUN echo "Downloading faster-whisper-small model..." && \
    modelscope download --model angelala00/faster-whisper-small --local_dir ./models/small --revision master || \
    (echo "faster-whisper download failed, retrying..." && \
     modelscope download --model angelala00/faster-whisper-small --local_dir ./models/small --revision master)

# 3. 下载 VoxCPM ONNX 模型
RUN echo "Downloading VoxCPM ONNX model..." && \
    aria2c -x 16 -s 16 --max-tries=5 --retry-wait=5 \
    -d models/ -o onnx_models_v15.zip \
    https://modelscope.cn/models/opqtest007/vox/resolve/master/onnx_models_v15.zip.ckpt || \
    (echo "VoxCPM download failed, retrying..." && \
     aria2c -x 16 -s 16 --max-tries=10 --retry-wait=10 \
     -d models/ -o onnx_models_v15.zip \
     https://modelscope.cn/models/opqtest007/vox/resolve/master/onnx_models_v15.zip.ckpt) && \
    cd models/ && \
    7z x onnx_models_v15.zip -p -y && \
    rm onnx_models_v15.zip && \
    cd ..

# 4. 下载 VoxCPM 参考音频数据库
RUN echo "Downloading VoxCPM reference database..." && \
    aria2c -x 16 -s 16 --max-tries=5 --retry-wait=5 \
    -d models/ -o voxcpm_ref.db \
    https://modelscope.cn/models/opqtest007/vox/resolve/master/voxcpm_ref.db || \
    (echo "VoxCPM reference database download failed, retrying..." && \
     aria2c -x 16 -s 16 --max-tries=10 --retry-wait=10 \
     -d models/ -o voxcpm_ref.db \
     https://modelscope.cn/models/opqtest007/vox/resolve/master/voxcpm_ref.db)

# 验证模型文件
RUN echo "Verifying model files..." && \
    ls -lh models/ && \
    ls -lh models/small/ 2>/dev/null || echo "small model not found" && \
    ls -lh models/onnx_models_v15/ 2>/dev/null || echo "onnx_models_v15 not found" && \
    ls -lh models/voxcpm_ref.db 2>/dev/null || echo "voxcpm_ref.db not found"

# 清理 pip 缓存和临时文件
RUN rm -rf /tmp/* /root/.cache/pip /root/.local/share/pip

# 暴露端口
EXPOSE 7860

# 创建非 root 用户（安全最佳实践）
RUN useradd -m -u 1000 appuser && \
    chown -R appuser:appuser /app

# 切换到非 root 用户
USER appuser

# 启动命令（使用 uvicorn 的生产配置）
# 注意：Hugging Face Spaces 会自动检测服务是否启动
CMD ["uvicorn", "app:app", \
     "--host", "0.0.0.0", \
     "--port", "7860", \
     "--workers", "1", \
     "--log-level", "info", \
     "--access-log", \
     "--no-use-colors", \
     "--timeout-keep-alive", "300"]