backend/Dockerfile.gpu

# BASE_REGISTRY 默认走 docker.io；国内可换 daocloud / 阿里云镜像（注意所选镜像需支持 nvidia/cuda 命名空间）
ARG BASE_REGISTRY=docker.io
FROM ${BASE_REGISTRY}/nvidia/cuda:12.4.1-cudnn-runtime-ubuntu22.04

ARG APT_MIRROR=mirrors.tuna.tsinghua.edu.cn
ARG PIP_INDEX=https://pypi.tuna.tsinghua.edu.cn/simple

RUN rm -f /etc/apt/sources.list && \
    rm -rf /etc/apt/sources.list.d/* && \
    echo "deb https://${APT_MIRROR}/ubuntu jammy main restricted universe multiverse" > /etc/apt/sources.list && \
    echo "deb https://${APT_MIRROR}/ubuntu jammy-updates main restricted universe multiverse" >> /etc/apt/sources.list && \
    echo "deb https://${APT_MIRROR}/ubuntu jammy-security main restricted universe multiverse" >> /etc/apt/sources.list && \
    apt-get update && \
    apt-get install -y --no-install-recommends ffmpeg python3-pip curl && \
    rm -rf /var/lib/apt/lists/*

ENV HF_ENDPOINT=https://hf-mirror.com

# 飞书「推送方式 = lark-cli / auto」时需要官方 lark CLI（npm 包 @larksuite/cli，二进制名 lark-cli）。
# Ubuntu 22.04 自带 apt 的 Node 太旧（v12）跑不动新 CLI，这里用 NodeSource 装 Node 20。
# 走 REST 直连推送则用不到，可按需删除本段以瘦身镜像。凭证由后端运行时经环境变量注入，不写死。
ARG NPM_REGISTRY=https://registry.npmmirror.com
RUN curl -fsSL https://deb.nodesource.com/setup_20.x | bash - && \
    apt-get install -y --no-install-recommends nodejs && \
    npm config set registry ${NPM_REGISTRY} && \
    npm install -g @larksuite/cli && \
    rm -rf /var/lib/apt/lists/* /root/.npm && \
    (lark-cli --version || true)

WORKDIR /app

# 先复制 requirements.txt 利用层缓存
COPY ./backend/requirements.txt /app/requirements.txt
RUN pip install --no-cache-dir -i ${PIP_INDEX} -r requirements.txt && \
    pip install --no-cache-dir -i ${PIP_INDEX} 'transformers[torch]>=4.23'

# 再复制应用代码
COPY ./backend /app

CMD ["python3", "main.py"]