a9867ee97c
- Introduced a new Dockerfile that builds the ConvertX-CN image with all necessary dependencies and models pre-downloaded. - Implemented multi-stage builds to optimize image size and build time. - Added a script for verifying the integrity of pre-downloaded models and dependencies. - Ensured the image operates in an offline-first manner, eliminating runtime dependencies on external networks.
653 lines
27 KiB
Text
653 lines
27 KiB
Text
# ==============================================================================
|
||
# ConvertX-CN 官方 Docker Image
|
||
# 版本：v0.1.11
|
||
# ==============================================================================
|
||
#
|
||
# 📦 Image 說明：
|
||
# - 這是 ConvertX-CN 官方 Docker Hub Image 的生產 Dockerfile
|
||
# - 已內建完整功能，無需額外擴充
|
||
# - ⚠️ 所有模型已在 build 階段預下載，runtime 不依賴網路
|
||
#
|
||
# 🌍 內建語言支援：
|
||
# - OCR: 英文、繁體中文、簡體中文、日文、韓文、德文、法文
|
||
# - Locale: en_US, zh_TW, zh_CN, ja_JP, ko_KR, de_DE, fr_FR
|
||
# - 字型: Noto CJK, Liberation, 標楷體
|
||
# - LaTeX: CJK、德文、法文、阿拉伯語、希伯來語
|
||
#
|
||
# 🤖 預下載模型清單：
|
||
# - PDFMathTranslate: DocLayout-YOLO ONNX（佈局分析）
|
||
# - BabelDOC: 資源透過顯式 Python 腳本下載，禁止 warmup
|
||
# - MinerU: PDF-Extract-Kit-1.0（Pipeline 模型）
|
||
# 包含：DocLayout-YOLO, YOLOv8 MFD, UniMERNet, PaddleOCR, LayoutReader, SLANet
|
||
#
|
||
# 📊 Image 大小：約 8-12 GB（含模型）
|
||
#
|
||
# ⚠️ Base Image：使用 debian:bookworm（穩定版）
|
||
# - 確保 Multi-Arch (amd64/arm64) 構建穩定性
|
||
# - 避免 trixie (testing) 套件同步不穩定問題
|
||
#
|
||
# 🔒 OFFLINE-FIRST 設計原則：
|
||
# 1. 所有下載行為只發生在 Docker build 階段
|
||
# 2. Runtime 完全不依賴外部網路
|
||
# 3. 禁止任何隱式下載（warmup、first-import、lazy-load）
|
||
# 4. 所有 cache 在同一 RUN 內清除，避免 layer 膨脹
|
||
#
|
||
# ==============================================================================
|
||
|
||
FROM debian:bookworm-slim AS base
|
||
LABEL org.opencontainers.image.source="https://github.com/pi-docket/ConvertX-CN"
|
||
LABEL org.opencontainers.image.description="ConvertX-CN - 精簡版檔案轉換服務"
|
||
LABEL org.opencontainers.image.version="0.1.11"
|
||
WORKDIR /app
|
||
|
||
# 配置 APT 重試機制（解決 Multi-Arch Build 時的網路不穩定問題）
|
||
RUN echo 'Acquire::Retries "5";' > /etc/apt/apt.conf.d/80-retries \
|
||
&& echo 'Acquire::http::Timeout "120";' >> /etc/apt/apt.conf.d/80-retries \
|
||
&& echo 'Acquire::https::Timeout "120";' >> /etc/apt/apt.conf.d/80-retries \
|
||
&& echo 'Acquire::ftp::Timeout "120";' >> /etc/apt/apt.conf.d/80-retries \
|
||
&& echo 'DPkg::Lock::Timeout "120";' >> /etc/apt/apt.conf.d/80-retries
|
||
|
||
# install bun
|
||
RUN apt-get update && apt-get install -y --no-install-recommends \
|
||
curl \
|
||
unzip \
|
||
ca-certificates \
|
||
&& rm -rf /var/lib/apt/lists/*
|
||
|
||
# if architecture is arm64, use the arm64 version of bun
|
||
RUN ARCH=$(uname -m) && \
|
||
if [ "$ARCH" = "aarch64" ]; then \
|
||
curl -fsSL -o bun-linux-aarch64.zip https://github.com/oven-sh/bun/releases/download/bun-v1.3.6/bun-linux-aarch64.zip; \
|
||
else \
|
||
curl -fsSL -o bun-linux-x64-baseline.zip https://github.com/oven-sh/bun/releases/download/bun-v1.3.6/bun-linux-x64-baseline.zip; \
|
||
fi
|
||
|
||
RUN unzip -j bun-linux-*.zip -d /usr/local/bin && \
|
||
rm bun-linux-*.zip && \
|
||
chmod +x /usr/local/bin/bun
|
||
|
||
# install dependencies into temp directory
|
||
# this will cache them and speed up future builds
|
||
FROM base AS install
|
||
RUN mkdir -p /temp/dev
|
||
COPY package.json bun.lock /temp/dev/
|
||
RUN cd /temp/dev && bun install --frozen-lockfile
|
||
|
||
# install with --production (exclude devDependencies)
|
||
RUN mkdir -p /temp/prod
|
||
COPY package.json bun.lock /temp/prod/
|
||
RUN cd /temp/prod && bun install --frozen-lockfile --production
|
||
|
||
FROM base AS prerelease
|
||
WORKDIR /app
|
||
COPY --from=install /temp/dev/node_modules node_modules
|
||
COPY . .
|
||
|
||
# ENV NODE_ENV=production
|
||
RUN bun run build
|
||
|
||
# copy production dependencies and source code into final image
|
||
FROM base AS release
|
||
|
||
# ==============================================================================
|
||
# 依賴安裝（分段安裝，優化 Multi-Arch Build 穩定性）
|
||
# ==============================================================================
|
||
#
|
||
# ✅ 核心轉換工具：完整保留
|
||
# ✅ TexLive：完整 CJK + 德法 + 阿拉伯/希伯來語
|
||
# ✅ OCR：7 種主要語言
|
||
# ✅ 字型：Noto CJK + Liberation + 標楷體
|
||
# ✅ OpenCV：電腦視覺轉換支援
|
||
# ✅ 額外影片編解碼器
|
||
# ✅ PDFMathTranslate：PDF 翻譯引擎
|
||
#
|
||
# 📝 分段安裝說明：
|
||
# - 將套件拆分為多個 RUN 層，避免 QEMU 模擬時記憶體不足
|
||
# - 每段安裝後清理 apt cache，減少中間層大小
|
||
# - 最終 squash 時會合併為單一層
|
||
#
|
||
# ==============================================================================
|
||
|
||
# 配置 APT 重試機制（解決 Multi-Arch Build 時的網路不穩定問題）
|
||
RUN echo 'Acquire::Retries "5";' > /etc/apt/apt.conf.d/80-retries \
|
||
&& echo 'Acquire::http::Timeout "120";' >> /etc/apt/apt.conf.d/80-retries \
|
||
&& echo 'Acquire::https::Timeout "120";' >> /etc/apt/apt.conf.d/80-retries \
|
||
&& echo 'Acquire::ftp::Timeout "120";' >> /etc/apt/apt.conf.d/80-retries \
|
||
&& echo 'APT::Get::Assume-Yes "true";' >> /etc/apt/apt.conf.d/80-retries \
|
||
&& echo 'DPkg::Lock::Timeout "120";' >> /etc/apt/apt.conf.d/80-retries
|
||
|
||
# 階段 1：基礎系統工具
|
||
RUN apt-get update --fix-missing && apt-get install -y --no-install-recommends \
|
||
locales \
|
||
ca-certificates \
|
||
curl \
|
||
&& rm -rf /var/lib/apt/lists/*
|
||
|
||
# 階段 2：核心轉換工具（小型）
|
||
# 注意：dasel 和 resvg 在 bookworm 中不存在，後續用二進位檔案安裝
|
||
RUN apt-get update --fix-missing && apt-get install -y --no-install-recommends \
|
||
assimp-utils \
|
||
dcraw \
|
||
dvisvgm \
|
||
ghostscript \
|
||
graphicsmagick \
|
||
mupdf-tools \
|
||
poppler-utils \
|
||
potrace \
|
||
&& rm -rf /var/lib/apt/lists/*
|
||
|
||
# 階段 2.1：安裝 dasel（從 GitHub 下載二進位檔案）
|
||
# ⬇️ Docker build 階段下載，runtime 不會再下載
|
||
RUN ARCH=$(uname -m) && \
|
||
if [ "$ARCH" = "aarch64" ]; then \
|
||
DASEL_ARCH="linux_arm64"; \
|
||
else \
|
||
DASEL_ARCH="linux_amd64"; \
|
||
fi && \
|
||
curl -sSLf "https://github.com/TomWright/dasel/releases/download/v2.8.1/dasel_${DASEL_ARCH}" -o /usr/local/bin/dasel && \
|
||
chmod +x /usr/local/bin/dasel
|
||
|
||
# 階段 2.2：安裝 resvg（從 GitHub 下載二進位檔案）
|
||
# 注意：resvg 官方只提供 x86_64 版本，ARM64 需從源碼編譯或跳過
|
||
# ⬇️ Docker build 階段下載，runtime 不會再下載
|
||
RUN ARCH=$(uname -m) && \
|
||
if [ "$ARCH" = "aarch64" ]; then \
|
||
echo "⚠️ resvg 沒有 ARM64 預編譯版本，跳過安裝（可改用 ImageMagick 或 Inkscape 替代）"; \
|
||
else \
|
||
curl -sSLf "https://github.com/linebender/resvg/releases/download/v0.44.0/resvg-linux-x86_64.tar.gz" -o /tmp/resvg.tar.gz && \
|
||
tar -xzf /tmp/resvg.tar.gz -C /tmp/ && \
|
||
mv /tmp/resvg /usr/local/bin/resvg && \
|
||
chmod +x /usr/local/bin/resvg && \
|
||
rm -rf /tmp/resvg.tar.gz; \
|
||
fi
|
||
|
||
# 階段 3：影音處理工具
|
||
RUN apt-get update --fix-missing && apt-get install -y --no-install-recommends \
|
||
ffmpeg \
|
||
libavcodec-extra \
|
||
libva2 \
|
||
&& rm -rf /var/lib/apt/lists/*
|
||
|
||
# 階段 4：圖像處理工具
|
||
# 注意：bookworm 使用 imagemagick（版本 6），trixie 才有 imagemagick-7
|
||
RUN apt-get update --fix-missing && apt-get install -y --no-install-recommends \
|
||
imagemagick \
|
||
inkscape \
|
||
libheif-examples \
|
||
libjxl-tools \
|
||
libvips-tools \
|
||
&& rm -rf /var/lib/apt/lists/*
|
||
|
||
# 階段 5：文件處理工具
|
||
RUN apt-get update --fix-missing && apt-get install -y --no-install-recommends \
|
||
calibre \
|
||
libemail-outlook-message-perl \
|
||
pandoc \
|
||
&& rm -rf /var/lib/apt/lists/*
|
||
|
||
# 階段 6：LibreOffice（最大的套件，單獨安裝）
|
||
RUN apt-get update --fix-missing && apt-get install -y --no-install-recommends \
|
||
libreoffice \
|
||
&& rm -rf /var/lib/apt/lists/*
|
||
|
||
# 階段 7：TexLive 基礎
|
||
RUN apt-get update --fix-missing && apt-get install -y --no-install-recommends \
|
||
texlive-base \
|
||
texlive-latex-base \
|
||
texlive-latex-recommended \
|
||
texlive-fonts-recommended \
|
||
texlive-xetex \
|
||
latexmk \
|
||
lmodern \
|
||
&& rm -rf /var/lib/apt/lists/*
|
||
|
||
# 階段 8：TexLive 語言包
|
||
RUN apt-get update --fix-missing && apt-get install -y --no-install-recommends \
|
||
texlive-lang-cjk \
|
||
texlive-lang-german \
|
||
texlive-lang-french \
|
||
texlive-lang-arabic \
|
||
texlive-lang-other \
|
||
&& rm -rf /var/lib/apt/lists/*
|
||
|
||
# 階段 9：OCR 支援
|
||
RUN apt-get update --fix-missing && apt-get install -y --no-install-recommends \
|
||
tesseract-ocr \
|
||
tesseract-ocr-eng \
|
||
tesseract-ocr-chi-tra \
|
||
tesseract-ocr-chi-sim \
|
||
tesseract-ocr-jpn \
|
||
tesseract-ocr-kor \
|
||
tesseract-ocr-deu \
|
||
tesseract-ocr-fra \
|
||
&& rm -rf /var/lib/apt/lists/*
|
||
|
||
# 階段 10：字型
|
||
RUN apt-get update --fix-missing && apt-get install -y --no-install-recommends \
|
||
fonts-noto-cjk \
|
||
fonts-noto-core \
|
||
fonts-noto-color-emoji \
|
||
fonts-liberation \
|
||
&& rm -rf /var/lib/apt/lists/*
|
||
|
||
# 階段 11：Python 依賴
|
||
RUN apt-get update --fix-missing && apt-get install -y --no-install-recommends \
|
||
python3 \
|
||
python3-pip \
|
||
python3-numpy \
|
||
python3-tinycss2 \
|
||
python3-opencv \
|
||
pipx \
|
||
&& rm -rf /var/lib/apt/lists/*
|
||
|
||
# ==============================================================================
|
||
# 🔧 Python 工具安裝策略
|
||
# ==============================================================================
|
||
#
|
||
# ⚠️ 重要原則：
|
||
# 1. pipx install 只安裝 CLI 工具本身
|
||
# 2. 模型/資源下載必須在後續單獨的 RUN 中顯式執行
|
||
# 3. 禁止依賴 --warmup 或 first-import 隱式下載
|
||
# 4. 所有 cache 必須在同一 RUN 內清除
|
||
#
|
||
# ==============================================================================
|
||
|
||
# 階段 12：安裝 Python 基礎工具
|
||
# ⬇️ 此步驟只安裝 huggingface_hub 用於後續模型下載
|
||
# runtime 不會再使用此套件下載任何資源
|
||
RUN pip3 install --no-cache-dir --break-system-packages huggingface_hub \
|
||
&& rm -rf /root/.cache/pip /tmp/*
|
||
|
||
# 階段 12-A：安裝 markitdown（文件轉換工具）
|
||
# 注意：markitdown[all] 可能依賴 transformers，但不會在 import 時下載模型
|
||
# ⬇️ 此步驟為 Docker build 階段安裝，runtime 不會再下載
|
||
RUN pipx install "markitdown[all]" \
|
||
&& rm -rf /root/.cache/pip /root/.local/pipx/.cache /tmp/*
|
||
|
||
# 階段 12-B：安裝 pdf2zh（PDFMathTranslate 引擎）
|
||
# 注意：pdf2zh 的模型將在後續階段顯式下載
|
||
# ⬇️ 此步驟為 Docker build 階段安裝，runtime 不會再下載
|
||
RUN pipx install "pdf2zh" \
|
||
&& rm -rf /root/.cache/huggingface /root/.cache/pip /root/.local/pipx/.cache /tmp/*
|
||
|
||
# 階段 12-C：安裝 babeldoc（BabelDOC 引擎）
|
||
# ⚠️ 注意：不使用 --warmup，資源將在後續階段顯式下載
|
||
# ⬇️ 此步驟為 Docker build 階段安裝，runtime 不會再下載
|
||
RUN (pipx install "babeldoc" || echo "⚠️ babeldoc 安裝失敗，跳過...") \
|
||
&& rm -rf /root/.cache/huggingface /root/.cache/pip /root/.local/pipx/.cache /tmp/*
|
||
|
||
# 階段 12-D：安裝 mineru（MinerU PDF 解析引擎）
|
||
# ⚠️ 注意：mineru[all] 依賴大量 HuggingFace 套件，但安裝時不應下載模型
|
||
# 模型下載將在後續階段顯式執行
|
||
# ⬇️ 此步驟為 Docker build 階段安裝，runtime 不會再下載
|
||
RUN (pipx install "mineru[all]" || echo "⚠️ mineru 安裝失敗（可能是 arm64 相容性問題），跳過...") \
|
||
&& rm -rf /root/.cache/huggingface /root/.cache/pip /root/.cache/torch /root/.local/pipx/.cache /tmp/*
|
||
|
||
# Add pipx bin directory to PATH（必須在模型下載前設定）
|
||
ENV PATH="/root/.local/bin:${PATH}"
|
||
|
||
# ==============================================================================
|
||
# 🔥 模型預下載區塊（Docker Build 階段）
|
||
# ==============================================================================
|
||
#
|
||
# ⚠️ 核心原則（OFFLINE-FIRST）：
|
||
# 1. 所有模型必須在 build 階段下載完成
|
||
# 2. Runtime 完全不依賴外部網路
|
||
# 3. 禁止任何隱式下載行為（warmup、lazy-load、first-import）
|
||
# 4. 使用顯式 Python 腳本下載，而非 CLI warmup
|
||
# 5. 所有 cache 必須在同一 RUN 內清除
|
||
#
|
||
# 📦 預下載的模型清單：
|
||
# 1. PDFMathTranslate / pdf2zh
|
||
# - DocLayout-YOLO ONNX 模型（佈局分析）
|
||
# 2. MinerU / magic-pdf
|
||
# - DocLayout-YOLO（佈局分析）
|
||
# - YOLOv8 MFD（公式偵測）
|
||
# - UniMERNet（公式辨識）
|
||
# - PaddleOCR（文字辨識）
|
||
# - LayoutReader（閱讀順序）
|
||
# - SLANet / UNet（表格辨識）
|
||
# 3. BabelDOC
|
||
# - 翻譯服務不需要本地模型（使用 API）
|
||
# - 字型和資源透過顯式下載
|
||
#
|
||
# ==============================================================================
|
||
# 🔧 BuildKit 優化說明：
|
||
# ==============================================================================
|
||
# 解決 "no space left on device" 的核心策略：
|
||
#
|
||
# 1. 【單一 RUN 原則】
|
||
# 所有模型下載 + cache 清理必須在同一個 RUN 中完成
|
||
# 這樣 BuildKit 在計算 layer diff 時，只會看到「最終狀態」
|
||
# 而不是「下載的 blob cache + 複製的模型」兩份資料
|
||
#
|
||
# 2. 【HuggingFace cache 必須刪除】
|
||
# snapshot_download 會在 ~/.cache/huggingface/hub 下建立：
|
||
# - blobs/：實際的模型檔案（用 SHA256 命名）
|
||
# - snapshots/：指向 blobs 的 symlink 或複製
|
||
# 當 local_dir_use_symlinks=False 時，檔案會被「複製」到目標目錄
|
||
# 如果不刪除 cache，同一份模型會以兩份大小進入 layer diff
|
||
#
|
||
# 3. 【避免 overlayfs 重複壓縮】
|
||
# exporting layers 時，BuildKit 會：
|
||
# - 計算每層的 diff（新增/修改的檔案）
|
||
# - 壓縮 diff 並寫入 /var/lib/buildkit/runc-overlayfs/
|
||
# 如果 cache 沒刪，diff 會包含 cache + 目標目錄，壓縮時空間翻倍
|
||
#
|
||
# ==============================================================================
|
||
|
||
# ------------------------------------------------------------------------------
|
||
# 階段 14-UNIFIED：所有模型下載 + 快取清理（單一 RUN 避免 layer 爆炸）
|
||
# ------------------------------------------------------------------------------
|
||
# 🔑 關鍵：這個 RUN 必須包含所有下載操作，並在結尾清理所有 cache
|
||
# 這樣 overlayfs 的 diff 只包含「最終需要的模型檔案」
|
||
# 而不是「cache 結構 + 模型副本」
|
||
#
|
||
# ⬇️ 以下為 Docker build 階段下載所有模型
|
||
# runtime 不會再下載任何資源
|
||
# ------------------------------------------------------------------------------
|
||
RUN set -eux && \
|
||
echo "===========================================================" && \
|
||
echo "🚀 開始統一模型下載（單一 RUN 優化 BuildKit layer）" && \
|
||
echo "===========================================================" && \
|
||
\
|
||
# ========================================
|
||
# [1/5] PDFMathTranslate DocLayout-YOLO ONNX 模型
|
||
# ========================================
|
||
# ⬇️ Docker build 階段下載 DocLayout-YOLO ONNX 模型
|
||
# 這是 pdf2zh 用於 PDF 版面分析的核心模型
|
||
# runtime 不會再下載此資源
|
||
echo "" && \
|
||
echo "📥 [1/5] 下載 DocLayout-YOLO ONNX 模型..." && \
|
||
mkdir -p /models/pdfmathtranslate && \
|
||
python3 -c " \
|
||
import os; \
|
||
os.environ['HF_HUB_DISABLE_PROGRESS_BARS'] = '0'; \
|
||
from huggingface_hub import snapshot_download; \
|
||
snapshot_download( \
|
||
repo_id='wybxc/DocLayout-YOLO-DocStructBench-onnx', \
|
||
local_dir='/models/pdfmathtranslate', \
|
||
allow_patterns=['*.onnx'], \
|
||
local_dir_use_symlinks=False \
|
||
)" && \
|
||
echo "✅ DocLayout-YOLO ONNX 下載完成" && \
|
||
ls -lh /models/pdfmathtranslate/*.onnx 2>/dev/null || ls -lh /models/pdfmathtranslate/ && \
|
||
\
|
||
# 🔥 立即清理 HuggingFace cache（關鍵！避免 blob 重複進入 layer）
|
||
rm -rf /root/.cache/huggingface && \
|
||
\
|
||
# ========================================
|
||
# [2/5] BabelDOC 資源（顯式下載，禁止 warmup）
|
||
# ========================================
|
||
# ⚠️ 重要：不使用 babeldoc --warmup，因為：
|
||
# 1. --warmup 內部使用 async HTTP，在 QEMU 下不穩定
|
||
# 2. --warmup 可能失敗但不報錯
|
||
# 3. 我們需要完全可控的下載流程
|
||
#
|
||
# BabelDOC 翻譯服務使用外部 API（Google/DeepL/OpenAI）
|
||
# 本地只需要字型和基礎資源，這些已透過 apt 安裝
|
||
# ⬇️ Docker build 階段準備 BabelDOC 快取目錄
|
||
# runtime 使用 API 翻譯，不需要本地模型
|
||
echo "" && \
|
||
echo "📥 [2/5] 準備 BabelDOC 資源..." && \
|
||
mkdir -p /root/.cache/babeldoc && \
|
||
if command -v babeldoc >/dev/null 2>&1; then \
|
||
echo "✅ BabelDOC 已安裝，翻譯將使用外部 API（Google/DeepL/OpenAI）"; \
|
||
echo " 本地不需要下載翻譯模型"; \
|
||
else \
|
||
echo "⚠️ BabelDOC 未安裝，跳過"; \
|
||
fi && \
|
||
echo "✅ BabelDOC 步驟完成" && \
|
||
\
|
||
# ========================================
|
||
# [3/5] PDFMathTranslate 多語言字型
|
||
# ========================================
|
||
# ⬇️ Docker build 階段下載 PDFMathTranslate 所需字型
|
||
# 這些字型用於 PDF 翻譯時保持正確的文字渲染
|
||
# runtime 不會再下載此資源
|
||
echo "" && \
|
||
echo "📥 [3/5] 下載 PDFMathTranslate 多語言字型..." && \
|
||
mkdir -p /app && \
|
||
curl -fSL --retry 3 --retry-delay 5 -o /app/GoNotoKurrent-Regular.ttf \
|
||
"https://github.com/satbyy/go-noto-universal/releases/download/v7.0/GoNotoKurrent-Regular.ttf" && \
|
||
curl -fSL --retry 3 --retry-delay 5 -o /app/SourceHanSerifCN-Regular.ttf \
|
||
"https://github.com/timelic/source-han-serif/releases/download/main/SourceHanSerifCN-Regular.ttf" && \
|
||
curl -fSL --retry 3 --retry-delay 5 -o /app/SourceHanSerifTW-Regular.ttf \
|
||
"https://github.com/timelic/source-han-serif/releases/download/main/SourceHanSerifTW-Regular.ttf" && \
|
||
curl -fSL --retry 3 --retry-delay 5 -o /app/SourceHanSerifJP-Regular.ttf \
|
||
"https://github.com/timelic/source-han-serif/releases/download/main/SourceHanSerifJP-Regular.ttf" && \
|
||
curl -fSL --retry 3 --retry-delay 5 -o /app/SourceHanSerifKR-Regular.ttf \
|
||
"https://github.com/timelic/source-han-serif/releases/download/main/SourceHanSerifKR-Regular.ttf" && \
|
||
echo "✅ 字型下載完成" && \
|
||
\
|
||
# ========================================
|
||
# [4/5] MinerU Pipeline 模型（顯式下載）
|
||
# ========================================
|
||
# ⬇️ Docker build 階段下載 MinerU Pipeline 模型
|
||
# 包含：DocLayout-YOLO, YOLOv8 MFD, UniMERNet, PaddleOCR, LayoutReader, SLANet
|
||
# runtime 不會再下載任何資源
|
||
echo "" && \
|
||
echo "📥 [4/5] 下載 MinerU Pipeline 模型..." && \
|
||
ARCH=$(uname -m) && \
|
||
MINERU_MODELS_DIR="/models/mineru" && \
|
||
mkdir -p "$MINERU_MODELS_DIR" && \
|
||
if [ "$ARCH" = "aarch64" ]; then \
|
||
echo "⚠️ ARM64 架構：MinerU 可能不完全支援，嘗試下載模型..."; \
|
||
fi && \
|
||
if command -v mineru >/dev/null 2>&1; then \
|
||
echo "使用顯式 Python 腳本下載 MinerU 模型..."; \
|
||
python3 -c " \
|
||
import os, json; \
|
||
os.environ['HF_HUB_DISABLE_PROGRESS_BARS'] = '0'; \
|
||
from huggingface_hub import snapshot_download; \
|
||
# PDF-Extract-Kit-1.0 Pipeline 模型
|
||
models_dir = '/models/mineru'; \
|
||
print('下載 PDF-Extract-Kit-1.0 模型...'); \
|
||
try: \
|
||
snapshot_download( \
|
||
repo_id='opendatalab/PDF-Extract-Kit-1.0', \
|
||
local_dir=os.path.join(models_dir, 'PDF-Extract-Kit-1.0'), \
|
||
local_dir_use_symlinks=False \
|
||
); \
|
||
print('✅ PDF-Extract-Kit-1.0 下載完成'); \
|
||
except Exception as e: \
|
||
print(f'⚠️ PDF-Extract-Kit-1.0 下載失敗: {e}'); \
|
||
# 建立 mineru.json 設定檔
|
||
config = { \
|
||
'models-dir': { \
|
||
'pipeline': os.path.join(models_dir, 'PDF-Extract-Kit-1.0'), \
|
||
'vlm': '' \
|
||
}, \
|
||
'model-source': 'local', \
|
||
'latex-delimiter-config': { \
|
||
'display': {'left': '\$\$', 'right': '\$\$'}, \
|
||
'inline': {'left': '\$', 'right': '\$'} \
|
||
} \
|
||
}; \
|
||
with open('/root/mineru.json', 'w') as f: \
|
||
json.dump(config, f, indent=2); \
|
||
print('✅ mineru.json 已建立'); \
|
||
" || echo "⚠️ MinerU 模型下載失敗，將在 runtime 時嘗試下載"; \
|
||
else \
|
||
echo "⚠️ mineru 命令不可用，跳過模型下載"; \
|
||
echo '{"models-dir":{"pipeline":"","vlm":""},"model-source":"huggingface","latex-delimiter-config":{"display":{"left":"$$","right":"$$"},"inline":{"left":"$","right":"$"}}}' > /root/mineru.json; \
|
||
fi && \
|
||
echo "✅ MinerU 模型下載步驟完成" && \
|
||
\
|
||
# 🔥 再次清理 HuggingFace cache（MinerU 也會產生）
|
||
rm -rf /root/.cache/huggingface && \
|
||
\
|
||
# ========================================
|
||
# [5/5] 最終驗證
|
||
# ========================================
|
||
echo "" && \
|
||
echo "📥 [5/5] 驗證模型完整性..." && \
|
||
echo "" && \
|
||
\
|
||
# ========================================
|
||
# 🔥 最終 Cache 清理（關鍵！避免 overlayfs diff 爆炸）
|
||
# ========================================
|
||
echo "===========================================================" && \
|
||
echo "🧹 清理所有下載快取（降低 layer diff 大小）" && \
|
||
echo "===========================================================" && \
|
||
# HuggingFace Hub cache（最大宗！包含所有 blob）
|
||
rm -rf /root/.cache/huggingface && \
|
||
# pip / Python build cache
|
||
rm -rf /root/.cache/pip && \
|
||
rm -rf /root/.cache/uv && \
|
||
# pipx cache
|
||
rm -rf /root/.local/pipx/.cache && \
|
||
# torch cache（可能由 MinerU 產生）
|
||
rm -rf /root/.cache/torch && \
|
||
# 通用 cache 目錄
|
||
rm -rf /tmp/* && \
|
||
rm -rf /var/tmp/* && \
|
||
# Python bytecode cache（可選，節省少量空間）
|
||
find /root/.local -type d -name "__pycache__" -exec rm -rf {} + 2>/dev/null || true && \
|
||
find /usr -type d -name "__pycache__" -exec rm -rf {} + 2>/dev/null || true && \
|
||
\
|
||
echo "" && \
|
||
echo "===========================================================" && \
|
||
echo "📋 模型檔案驗證" && \
|
||
echo "===========================================================" && \
|
||
echo "" && \
|
||
echo "🔹 PDFMathTranslate 模型：" && \
|
||
ONNX_COUNT=$(find /models/pdfmathtranslate -name "*.onnx" 2>/dev/null | wc -l) && \
|
||
if [ "$ONNX_COUNT" -gt 0 ]; then \
|
||
echo " ✅ 找到 $ONNX_COUNT 個 ONNX 模型："; \
|
||
ls -lh /models/pdfmathtranslate/*.onnx 2>/dev/null || find /models/pdfmathtranslate -name "*.onnx" -exec ls -lh {} \;; \
|
||
else \
|
||
echo " ❌ /models/pdfmathtranslate 中沒有 ONNX 模型"; \
|
||
fi && \
|
||
echo "" && \
|
||
echo "🔹 PDFMathTranslate 字型：" && \
|
||
ls -lh /app/*.ttf 2>/dev/null || echo " ⚠️ 無字型檔案" && \
|
||
echo "" && \
|
||
echo "🔹 BabelDOC 狀態：" && \
|
||
if command -v babeldoc >/dev/null 2>&1; then \
|
||
echo " ✅ BabelDOC 已安裝（使用外部 API 翻譯）"; \
|
||
else \
|
||
echo " ⚠️ BabelDOC 未安裝"; \
|
||
fi && \
|
||
echo "" && \
|
||
echo "🔹 MinerU 模型目錄：" && \
|
||
if [ -d "/models/mineru/PDF-Extract-Kit-1.0" ]; then \
|
||
echo " ✅ MinerU Pipeline 模型已下載"; \
|
||
du -sh /models/mineru/PDF-Extract-Kit-1.0 2>/dev/null || true; \
|
||
else \
|
||
echo " ⚠️ MinerU 模型目錄不存在（可能需要 runtime 下載）"; \
|
||
fi && \
|
||
echo "" && \
|
||
echo "🔹 mineru.json 設定：" && \
|
||
if [ -f /root/mineru.json ]; then \
|
||
cat /root/mineru.json; \
|
||
else \
|
||
echo " ⚠️ mineru.json 不存在"; \
|
||
fi && \
|
||
echo "" && \
|
||
echo "🔹 確認 HuggingFace cache 已清除：" && \
|
||
if [ -d "/root/.cache/huggingface" ]; then \
|
||
echo " ❌ 警告：HuggingFace cache 仍存在！"; \
|
||
du -sh /root/.cache/huggingface 2>/dev/null || true; \
|
||
else \
|
||
echo " ✅ HuggingFace cache 已清除"; \
|
||
fi && \
|
||
echo "" && \
|
||
echo "===========================================================" && \
|
||
echo "✅ 模型下載完成，所有快取已清理" && \
|
||
echo "==========================================================="
|
||
|
||
# ==============================================================================
|
||
# 環境變數設定
|
||
# ==============================================================================
|
||
|
||
# PDFMathTranslate 環境變數
|
||
ENV PDFMATHTRANSLATE_MODELS_PATH="/models/pdfmathtranslate"
|
||
ENV NOTO_FONT_PATH="/app/GoNotoKurrent-Regular.ttf"
|
||
|
||
# BabelDOC 環境變數
|
||
ENV BABELDOC_CACHE_PATH="/root/.cache/babeldoc"
|
||
ENV BABELDOC_SERVICE="google"
|
||
|
||
# MinerU 環境變數
|
||
# 強制使用本地模型，禁止 runtime 下載
|
||
ENV MINERU_MODELS_PATH="/models/mineru"
|
||
|
||
# HuggingFace 離線模式（禁止 runtime 下載）
|
||
# ⚠️ 這是 OFFLINE-FIRST 設計的核心
|
||
ENV HF_HUB_OFFLINE="1"
|
||
ENV TRANSFORMERS_OFFLINE="1"
|
||
|
||
# ==============================================================================
|
||
# 最終清理（模型下載完成後）
|
||
# ==============================================================================
|
||
RUN rm -rf /usr/share/doc/texlive* \
|
||
&& rm -rf /usr/share/texlive/texmf-dist/doc \
|
||
&& rm -rf /usr/share/doc/* \
|
||
&& rm -rf /usr/share/man/* \
|
||
&& rm -rf /usr/share/info/*
|
||
|
||
# ==============================================================================
|
||
# 設定 locale（支援中文 PDF 避免亂碼）
|
||
# ==============================================================================
|
||
RUN sed -i 's/# en_US.UTF-8 UTF-8/en_US.UTF-8 UTF-8/' /etc/locale.gen && \
|
||
sed -i 's/# zh_TW.UTF-8 UTF-8/zh_TW.UTF-8 UTF-8/' /etc/locale.gen && \
|
||
sed -i 's/# zh_CN.UTF-8 UTF-8/zh_CN.UTF-8 UTF-8/' /etc/locale.gen && \
|
||
sed -i 's/# ja_JP.UTF-8 UTF-8/ja_JP.UTF-8 UTF-8/' /etc/locale.gen && \
|
||
sed -i 's/# ko_KR.UTF-8 UTF-8/ko_KR.UTF-8 UTF-8/' /etc/locale.gen && \
|
||
sed -i 's/# de_DE.UTF-8 UTF-8/de_DE.UTF-8 UTF-8/' /etc/locale.gen && \
|
||
sed -i 's/# fr_FR.UTF-8 UTF-8/fr_FR.UTF-8 UTF-8/' /etc/locale.gen && \
|
||
locale-gen
|
||
|
||
# 預設使用 zh_TW.UTF-8 確保中文 PDF 正確顯示
|
||
ENV LANG=zh_TW.UTF-8
|
||
ENV LC_ALL=zh_TW.UTF-8
|
||
|
||
# ==============================================================================
|
||
# 安裝自訂字型（標楷體等台灣常用字型）
|
||
# ==============================================================================
|
||
RUN mkdir -p /usr/share/fonts/truetype/custom
|
||
COPY fonts/ /usr/share/fonts/truetype/custom/
|
||
RUN fc-cache -fv
|
||
|
||
# ==============================================================================
|
||
# Install VTracer binary（向量追蹤工具）
|
||
# ⬇️ Docker build 階段下載，runtime 不會再下載
|
||
# ==============================================================================
|
||
RUN ARCH=$(uname -m) && \
|
||
if [ "$ARCH" = "aarch64" ]; then \
|
||
VTRACER_ASSET="vtracer-aarch64-unknown-linux-musl.tar.gz"; \
|
||
else \
|
||
VTRACER_ASSET="vtracer-x86_64-unknown-linux-musl.tar.gz"; \
|
||
fi && \
|
||
curl -L --retry 3 --retry-delay 5 -o /tmp/vtracer.tar.gz "https://github.com/visioncortex/vtracer/releases/download/0.6.4/${VTRACER_ASSET}" && \
|
||
tar -xzf /tmp/vtracer.tar.gz -C /tmp/ && \
|
||
mv /tmp/vtracer /usr/local/bin/vtracer && \
|
||
chmod +x /usr/local/bin/vtracer && \
|
||
rm /tmp/vtracer.tar.gz
|
||
|
||
COPY --from=install /temp/prod/node_modules node_modules
|
||
COPY --from=prerelease /app/public/ /app/public/
|
||
COPY --from=prerelease /app/dist /app/dist
|
||
|
||
# 複製模型驗證腳本
|
||
COPY scripts/verify-models.sh /app/scripts/verify-models.sh
|
||
RUN chmod +x /app/scripts/verify-models.sh
|
||
|
||
RUN mkdir data
|
||
|
||
EXPOSE 3000/tcp
|
||
|
||
# ==============================================================================
|
||
# 環境變數
|
||
# ==============================================================================
|
||
# Calibre 需要
|
||
ENV QTWEBENGINE_CHROMIUM_FLAGS="--no-sandbox"
|
||
# Pandoc PDF 引擎（使用 pdflatex 以獲得最佳相容性）
|
||
ENV PANDOC_PDF_ENGINE=pdflatex
|
||
# Node 環境
|
||
ENV NODE_ENV=production
|
||
# PDFMathTranslate 預設翻譯服務（可透過環境變數覆寫）
|
||
ENV PDFMATHTRANSLATE_SERVICE="google"
|
||
|
||
ENTRYPOINT [ "bun", "run", "dist/src/index.js" ]
|