diff --git a/Dockerfile.v0.1.11 b/Dockerfile.v0.1.11 index d3a6276..4256eab 100644 --- a/Dockerfile.v0.1.11 +++ b/Dockerfile.v0.1.11 @@ -42,29 +42,29 @@ 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 + && 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/* + 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 + 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 + rm bun-linux-*.zip && \ + chmod +x /usr/local/bin/bun # install dependencies into temp directory # this will cache them and speed up future builds @@ -110,135 +110,135 @@ FROM base AS release # 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 + && 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/* + 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/* + 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 + 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 + 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/* + 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/* + 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/* + 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/* + 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/* + 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/* + 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/* + 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/* + 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/* + python3 \ + python3-pip \ + python3-numpy \ + python3-tinycss2 \ + python3-opencv \ + pipx \ + && rm -rf /var/lib/apt/lists/* # ============================================================================== # 🔧 Python 工具安裝策略 @@ -256,32 +256,32 @@ RUN apt-get update --fix-missing && apt-get install -y --no-install-recommends \ # ⬇️ 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/* + && 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/* + && 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/* + && 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/* + && 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/* + && 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}" @@ -347,215 +347,215 @@ ENV PATH="/root/.local/bin:${PATH}" # 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( \ + 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 "✅ 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 \ + 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 \ + 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 \ + 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 \ + 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 \ + 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 \ + 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 "✅ 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 \ + 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: \ + 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 \ + 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: \ + except Exception as e: \ print(f'⚠️ PDF-Extract-Kit-1.0 下載失敗: {e}'); \ -# 建立 mineru.json 設定檔 -config = { \ + # 建立 mineru.json 設定檔 + config = { \ 'models-dir': { \ - 'pipeline': os.path.join(models_dir, 'PDF-Extract-Kit-1.0'), \ - 'vlm': '' \ + 'pipeline': os.path.join(models_dir, 'PDF-Extract-Kit-1.0'), \ + 'vlm': '' \ }, \ 'model-source': 'local', \ 'latex-delimiter-config': { \ - 'display': {'left': '\$\$', 'right': '\$\$'}, \ - 'inline': {'left': '\$', 'right': '\$'} \ + 'display': {'left': '\$\$', 'right': '\$\$'}, \ + 'inline': {'left': '\$', 'right': '\$'} \ } \ -}; \ -with open('/root/mineru.json', 'w') as f: \ + }; \ + 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 \ + 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 \ + 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 \ + 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 \ + 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 \ + else \ echo " ⚠️ BabelDOC 未安裝"; \ - fi && \ - echo "" && \ - echo "🔹 MinerU 模型目錄：" && \ - if [ -d "/models/mineru/PDF-Extract-Kit-1.0" ]; then \ + 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 \ + 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 \ + fi && \ + echo "" && \ + echo "🔹 mineru.json 設定：" && \ + if [ -f /root/mineru.json ]; then \ cat /root/mineru.json; \ - else \ + else \ echo " ⚠️ mineru.json 不存在"; \ - fi && \ - echo "" && \ - echo "🔹 確認 HuggingFace cache 已清除：" && \ - if [ -d "/root/.cache/huggingface" ]; then \ + 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 \ + 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 "===========================================================" + fi && \ + echo "" && \ + echo "===========================================================" && \ + echo "✅ 模型下載完成，所有快取已清理" && \ + echo "===========================================================" # ============================================================================== # 環境變數設定 @@ -582,22 +582,22 @@ 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/* + && 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 + 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 @@ -615,16 +615,16 @@ RUN fc-cache -fv # ⬇️ 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 + 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/