feat: Introduce modular docling-slim package (#3285)

* plans folder structure Signed-off-by: Michele Dolfi <dol@zurich.ibm.com> * initial plan Signed-off-by: Michele Dolfi <dol@zurich.ibm.com> * updated plan Signed-off-by: Michele Dolfi <dol@zurich.ibm.com> * restructure repo for docling and docling-slim Signed-off-by: Michele Dolfi <dol@zurich.ibm.com> * transpose package structures Signed-off-by: Michele Dolfi <dol@zurich.ibm.com> * add all-packages Signed-off-by: Michele Dolfi <dol@zurich.ibm.com> * updated lock and deps Signed-off-by: Michele Dolfi <dol@zurich.ibm.com> * align deps Signed-off-by: Michele Dolfi <dol@zurich.ibm.com> * more lock like main Signed-off-by: Michele Dolfi <dol@zurich.ibm.com> * more locked pinning Signed-off-by: Michele Dolfi <dol@zurich.ibm.com> * rename extras Signed-off-by: Michele Dolfi <dol@zurich.ibm.com> * add simple README for docling-slim Signed-off-by: Michele Dolfi <dol@zurich.ibm.com> * fix scikit-image issue Signed-off-by: Michele Dolfi <dol@zurich.ibm.com> * add readme placeholder Signed-off-by: Michele Dolfi <dol@zurich.ibm.com> * add all extras in package test Signed-off-by: Michele Dolfi <dol@zurich.ibm.com> * cli in docling-slim Signed-off-by: Michele Dolfi <dol@zurich.ibm.com> * apply formatting Signed-off-by: Michele Dolfi <dol@zurich.ibm.com> * fix testing package Signed-off-by: Michele Dolfi <dol@zurich.ibm.com> * override grpcio in no-header test Signed-off-by: Michele Dolfi <dol@zurich.ibm.com> * update lock Signed-off-by: Michele Dolfi <dol@zurich.ibm.com> * update package description Signed-off-by: Michele Dolfi <dol@zurich.ibm.com> * updated extras Signed-off-by: Michele Dolfi <dol@zurich.ibm.com> * fix publish scripts Signed-off-by: Michele Dolfi <dol@zurich.ibm.com> * update package test Signed-off-by: Michele Dolfi <dol@zurich.ibm.com> --------- Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>
2026-05-17 13:10:38 +00:00 · 2026-04-24 15:14:57 +02:00
parent a6a37ca895
commit ed32c5e993
17 changed files with 3490 additions and 1543 deletions
@@ -0,0 +1,27 @@
+#!/bin/bash
+
+set -e  # trigger failure on error - do not remove!
+set -x  # display command on output
+
+# Build each package into its own dist subdirectory so the PyPI publish
+# action can upload them independently (otherwise a single `dist/` causes
+# the second publish step to re-upload files and fail on `skip-existing: false`).
+
+# Build docling-slim package (from repo root — source co-located)
+echo "Building docling-slim package..."
+uv build --out-dir dist/docling-slim
+
+# Build docling package (meta-package, dependency-only wheel)
+echo "Building docling package..."
+# Backup placeholder README and copy root README for build
+mv packages/docling/README.md packages/docling/README.md.placeholder
+cp README.md packages/docling/README.md
+(cd packages/docling && uv build --out-dir ../../dist/docling)
+# Restore placeholder README
+mv packages/docling/README.md.placeholder packages/docling/README.md
+
+echo "Build complete."
+echo "docling-slim artifacts:"
+ls -lh dist/docling-slim/
+echo "docling artifacts:"
+ls -lh dist/docling/
@@ -9,9 +9,21 @@ if [ -z "${TARGET_VERSION}" ]; then
 fi
 CHGLOG_FILE="${CHGLOG_FILE:-CHANGELOG.md}"

-# update package version
+# update package versions:
+#   - root pyproject.toml  = docling-slim
+#   - packages/docling/pyproject.toml = docling (meta-package)
 uvx --from=toml-cli toml set --toml-path=pyproject.toml project.version "${TARGET_VERSION}"
-UV_FROZEN=0 uv lock --upgrade-package docling
+uvx --from=toml-cli toml set --toml-path=packages/docling/pyproject.toml project.version "${TARGET_VERSION}"
+
+# update docling-slim dependency version in docling package
+uvx --from=toml-cli toml set --toml-path=packages/docling/pyproject.toml "project.dependencies[0]" "docling-slim[standard]==${TARGET_VERSION}"
+
+# update all re-exported extras in docling package
+for extra in easyocr tesserocr ocrmac vlm rapidocr asr htmlrender remote-serving onnxruntime xbrl; do
+    uvx --from=toml-cli toml set --toml-path=packages/docling/pyproject.toml "project.optional-dependencies.${extra}[0]" "docling-slim[*]==${TARGET_VERSION}"
+done
+
+UV_FROZEN=0 uv lock --upgrade-package docling --upgrade-package docling-slim

 # collect release notes
 REL_NOTES=$(mktemp)
@@ -31,7 +43,7 @@ mv "${TMP_CHGLOG}" "${CHGLOG_FILE}"
 # push changes
 git config --global user.name 'github-actions[bot]'
 git config --global user.email 'github-actions[bot]@users.noreply.github.com'
-git add pyproject.toml uv.lock "${CHGLOG_FILE}"
+git add pyproject.toml packages/docling/pyproject.toml uv.lock "${CHGLOG_FILE}"
 COMMIT_MSG="chore: bump version to ${TARGET_VERSION} [skip ci]"
 git commit -m "${COMMIT_MSG}"
 git push origin main
@@ -50,7 +50,7 @@ jobs:
            pre-commit|${{ env.PY }}|

      - name: Install Python Dependencies
-        run: uv sync --frozen --all-extras
+        run: uv sync --frozen --all-extras --all-packages

      - name: Check style
        run: |
@@ -92,7 +92,7 @@ jobs:
          python-version: ${{ matrix.python-version }}

      - name: Install Python Dependencies
-        run: uv sync --frozen --all-extras
+        run: uv sync --frozen --all-extras --all-packages

      - name: Cache Models
        uses: actions/cache@v5
@@ -159,7 +159,7 @@ jobs:
          python-version: ${{ matrix.python-version }}

      - name: Install Python Dependencies
-        run: uv sync --frozen --all-extras
+        run: uv sync --frozen --all-extras --all-packages

      - name: Cache Models
        uses: actions/cache@v5
@@ -231,7 +231,7 @@ jobs:
          python-version: ${{ matrix.python-version }}

      - name: Install Python Dependencies
-        run: uv sync --frozen --all-extras
+        run: uv sync --frozen --all-extras --all-packages

      - name: Cache Models
        uses: actions/cache@v5
@@ -301,21 +301,21 @@ jobs:
            echo "=========================================="
            echo "Testing Python $py_version"
            echo "=========================================="
-            
+
            # Create virtual environment with uv
            uv venv /tmp/venv-${py_version} --python=${py_version}
            source /tmp/venv-${py_version}/bin/activate
-            
-            # Install package with pip (no lock file)
-            uv pip install --torch-backend=cpu -e .[easyocr,tesserocr,vlm,rapidocr,asr]
-            
+
+            # Install docling-slim with pip (no lock file)
+            uv pip install --torch-backend=cpu -e .[all]
+
            # Run basic import test
-            python -c "import docling; from docling.document_converter import DocumentConverter; print('Import successful for Python ${py_version}')"
-            
+            python -c "import docling; print('Import successful for Python ${py_version}')"
+
            # Cleanup
            deactivate
            rm -rf /tmp/venv-${py_version}
-            
+
            echo "Python $py_version: PASSED"
            echo ""
          done
@@ -334,16 +334,16 @@ jobs:
            echo "=========================================="
            echo "Testing Python $py_version (no dev headers)"
            echo "=========================================="
-            
+
            # Create virtual environment with uv
            uv venv /tmp/venv-nodev-${py_version} --python=${py_version}
            source /tmp/venv-nodev-${py_version}/bin/activate
-            
+
            # Find and remove Python.h from the Python installation
            echo "Removing Python development headers from Python installation..."
            python_include_dir=$(python -c "import sysconfig; print(sysconfig.get_path('include'))")
            echo "Python include directory: $python_include_dir"
-            
+
            if [ -f "$python_include_dir/Python.h" ]; then
              echo "Found Python.h, removing it and other headers..."
              # Use sudo if the directory is system-owned
@@ -356,7 +356,7 @@ jobs:
            else
              echo "Warning: Python.h not found at expected location"
            fi
-            
+
            # Verify that compilation fails without dev headers
            # Try to install numpy from source (sdist) - this should fail
            echo "Verifying compilation fails without dev headers..."
@@ -364,7 +364,7 @@ jobs:
            uv pip install --no-binary=:all: numpy==1.26.4 > /tmp/numpy-install-${py_version}.log 2>&1
            numpy_exit_code=$?
            set -e  # Re-enable exit on error
-            
+
            if [ $numpy_exit_code -eq 0 ]; then
              echo "ERROR: numpy installation from source succeeded, but it should have failed without dev headers!"
              cat /tmp/numpy-install-${py_version}.log
@@ -378,19 +378,18 @@ jobs:
                echo "Warning: Error message doesn't explicitly mention missing headers, but compilation failed as expected"
              fi
            fi
-            
-            # Install package with pip (no lock file, no compilation)
-            # Install without extras that require compilation (tesserocr requires dev headers)
-            # Note: Not using --only-binary since some packages are sdist-only but don't require compilation
-            uv pip install --torch-backend=cpu -e .[easyocr,vlm,rapidocr,asr]
-            
+
+            # Install docling-slim with pip (no lock file, no compilation)
+            echo "grpcio>=1.71.0" > override-grpcio.txt
+            uv pip install --torch-backend=cpu -e ".[all]" --overrides override-grpcio.txt
+
            # Run basic import test
-            python -c "import docling; from docling.document_converter import DocumentConverter; print('Import successful for Python ${py_version} without dev headers')"
-            
+            python -c "import docling; print('Import successful for Python ${py_version} without dev headers')"
+
            # Cleanup
            deactivate
            rm -rf /tmp/venv-nodev-${py_version}
-            
+
            echo "Python $py_version (no dev headers): PASSED"
            echo ""
          done
@@ -412,11 +411,15 @@ jobs:
      - name: Install dependencies
        run: uv sync --all-extras

-      - name: Build package
-        run: uv build
+      - name: Build packages
+        run: bash .github/scripts/build-packages.sh

-      - name: Check content of wheel
-        run: unzip -l dist/*.whl
+      - name: Check content of wheels
+        run: |
+          for whl in dist/*/*.whl; do
+            echo "=== $whl ==="
+            unzip -l "$whl"
+          done

      - name: Store the distribution packages
        uses: actions/upload-artifact@v6
@@ -447,7 +450,7 @@ jobs:

      - name: Install package
        run: |
-          uv pip install dist/*.whl
+          uv pip install --find-links dist/docling-slim/ dist/docling/docling-*.whl

      - name: Run docling
        run: uv run docling --help
@@ -19,8 +19,10 @@ jobs:
          with:
            python-version: ${{ matrix.python-version }}
            enable-cache: true
+        - name: Install Python Dependencies
+          run: uv sync --frozen --all-extras --all-packages
        - name: Build docs
-          run: uv run mkdocs build --verbose --clean
+          run: uv run --no-sync mkdocs build --verbose --clean
        - name: Build and push docs
          if: inputs.deploy
          run: uv run --no-sync mkdocs gh-deploy --force
@@ -1,4 +1,4 @@
-name: "Build and publish package"
+name: "Build and publish packages"

 on:
  release:
@@ -11,16 +11,11 @@ permissions:
  contents: read

 jobs:
-  build-and-publish:
+  build:
    runs-on: ubuntu-latest
    strategy:
      matrix:
        python-version: ['3.12']
-    environment:
-      name: pypi
-      url: https://pypi.org/p/docling
-    permissions:
-      id-token: write  # IMPORTANT: mandatory for trusted publishing
    steps:
      - uses: actions/checkout@v6
      - name: Install uv and set the python version
@@ -30,9 +25,58 @@ jobs:
          enable-cache: true
      - name: Install dependencies
        run: uv sync --all-extras
-      - name: Build package
-        run: uv build
-      - name: Publish distribution 📦 to PyPI
+      - name: Build packages
+        run: bash .github/scripts/build-packages.sh
+      - name: Upload build artifacts
+        uses: actions/upload-artifact@v4
+        with:
+          name: dist
+          path: dist/
+
+  publish-docling-slim:
+    needs: build
+    runs-on: ubuntu-latest
+    environment:
+      name: pypi
+      url: https://pypi.org/p/docling-slim
+    permissions:
+      id-token: write  # IMPORTANT: mandatory for trusted publishing
+    steps:
+      - name: Download build artifacts
+        uses: actions/download-artifact@v4
+        with:
+          name: dist
+          path: dist/
+      - name: Publish docling-slim to PyPI
        uses: pypa/gh-action-pypi-publish@release/v1
        with:
          attestations: true
+          packages-dir: dist/docling-slim/
+          skip-existing: false
+
+  publish-docling:
+    # docling is a meta-package that depends on docling-slim, so publish it
+    # after docling-slim is available on PyPI.
+    needs: publish-docling-slim
+    runs-on: ubuntu-latest
+    environment:
+      name: pypi
+      url: https://pypi.org/p/docling
+    permissions:
+      id-token: write  # IMPORTANT: mandatory for trusted publishing
+    steps:
+      - name: Download build artifacts
+        uses: actions/download-artifact@v4
+        with:
+          name: dist
+          path: dist/
+      - name: Wait for docling-slim to be available on PyPI
+        run: |
+          echo "Waiting 60 seconds for docling-slim to propagate on PyPI..."
+          sleep 60
+      - name: Publish docling to PyPI
+        uses: pypa/gh-action-pypi-publish@release/v1
+        with:
+          attestations: true
+          packages-dir: dist/docling/
+          skip-existing: false
@@ -1,6 +1,7 @@
 import datetime
 import logging
 import re
+import sys
 import tempfile
 import time
 import warnings
@@ -8,8 +9,25 @@ from collections.abc import Iterable
 from pathlib import Path
 from typing import Annotated, Type

-import rich.table
-import typer
+# Check for CLI dependencies
+try:
+    import rich.table
+    import typer
+except ImportError as e:
+    missing_package = str(e).split("'")[1] if "'" in str(e) else "typer or rich"
+    print(
+        f"Error: Missing required CLI dependency '{missing_package}'", file=sys.stderr
+    )
+    print("\nThe docling CLI requires additional dependencies.", file=sys.stderr)
+    print("Please install them using one of the following options:\n", file=sys.stderr)
+    print("  1. Install the full docling package (recommended):", file=sys.stderr)
+    print("     pip install docling\n", file=sys.stderr)
+    print("  2. Install docling-slim with CLI support:", file=sys.stderr)
+    print("     pip install docling-slim[cli]\n", file=sys.stderr)
+    print("  3. Install just the missing dependencies:", file=sys.stderr)
+    print("     pip install typer rich\n", file=sys.stderr)
+    sys.exit(1)
+
 from docling_core.transforms.serializer.html import (
    HTMLDocSerializer,
    HTMLOutputStyle,
@@ -1,12 +1,29 @@
 import logging
+import sys
 import warnings
 from enum import Enum
 from pathlib import Path
 from typing import Annotated, Optional

-import typer
-from rich.console import Console
-from rich.logging import RichHandler
+# Check for CLI dependencies
+try:
+    import typer
+    from rich.console import Console
+    from rich.logging import RichHandler
+except ImportError as e:
+    missing_package = str(e).split("'")[1] if "'" in str(e) else "typer or rich"
+    print(
+        f"Error: Missing required CLI dependency '{missing_package}'", file=sys.stderr
+    )
+    print("\nThe docling-tools CLI requires additional dependencies.", file=sys.stderr)
+    print("Please install them using one of the following options:\n", file=sys.stderr)
+    print("  1. Install the full docling package (recommended):", file=sys.stderr)
+    print("     pip install docling\n", file=sys.stderr)
+    print("  2. Install docling-slim with CLI support:", file=sys.stderr)
+    print("     pip install docling-slim[cli]\n", file=sys.stderr)
+    print("  3. Install just the missing dependencies:", file=sys.stderr)
+    print("     pip install typer rich\n", file=sys.stderr)
+    sys.exit(1)

 from docling.datamodel.settings import settings
 from docling.models.utils.hf_model_download import download_hf_model
@@ -1,4 +1,22 @@
-import typer
+import sys
+
+# Check for CLI dependencies
+try:
+    import typer
+except ImportError as e:
+    missing_package = str(e).split("'")[1] if "'" in str(e) else "typer"
+    print(
+        f"Error: Missing required CLI dependency '{missing_package}'", file=sys.stderr
+    )
+    print("\nThe docling-tools CLI requires additional dependencies.", file=sys.stderr)
+    print("Please install them using one of the following options:\n", file=sys.stderr)
+    print("  1. Install the full docling package (recommended):", file=sys.stderr)
+    print("     pip install docling\n", file=sys.stderr)
+    print("  2. Install docling-slim with CLI support:", file=sys.stderr)
+    print("     pip install docling-slim[cli]\n", file=sys.stderr)
+    print("  3. Install just the missing dependencies:", file=sys.stderr)
+    print("     pip install typer rich\n", file=sys.stderr)
+    sys.exit(1)

 from docling.cli.models import app as models_app

@@ -0,0 +1,126 @@
+# Docling Slim
+
+**Lightweight SDK for parsing documents with minimal dependencies and opt-in extras**
+
+Docling Slim is a minimal-dependency version of Docling that allows you to install only the components you need. It provides the core document processing functionality with ~50MB of base dependencies, and you can add specific features through optional extras.
+
+## When to Use Docling Slim
+
+- **Use `docling`** (recommended): If you want the full-featured experience with all standard capabilities
+- **Use `docling-slim`**: If you need fine-grained control over dependencies or want to minimize installation size
+
+## For Most Users: Use the Main Docling Package
+
+We recommend most users install the full-featured `docling` package instead:
+
+```bash
+pip install docling
+```
+
+The `docling` package includes all standard features, the CLI tools, and is the easiest way to get started. Visit the [main Docling documentation](https://docling-project.github.io/docling/) for complete guides and examples.
+
+## Installation
+
+### With Specific Features
+```bash
+# PDF support with local models
+pip install docling-slim[format-pdf,models-local]
+
+# Office formats only
+pip install docling-slim[format-office]
+
+# PDF + CLI
+pip install docling-slim[format-pdf,cli]
+
+# Docling service client for using the Docling Serve API
+pip install docling-slim[service-client]
+```
+
+## Available Extras
+
+### Convenience Bundles
+
+| Extra | Description | Use Case |
+|-------|-------------|----------|
+| `standard` | All standard features (same as `docling` package) | Full-featured usage |
+| `all` | All available extras | Complete installation |
+
+### CLI
+
+| Extra | Description | Use Case |
+|-------|-------------|----------|
+| `cli` | Command-line interface (typer, rich) | CLI tools (docling, docling-tools) |
+
+### Core Components
+
+| Extra | Description | Use Case |
+|-------|-------------|----------|
+| `convert-core` | Core conversion components (numpy, pillow, scipy) | Basic document conversion |
+| `extract-core` | Structured information extraction | Data extraction from documents |
+
+### Format Support
+
+#### PDF Formats
+
+| Extra | Description | Use Case |
+|-------|-------------|----------|
+| `format-pdf` | PDF parsing (pypdfium2 + docling-parse) | PDF documents |
+| `format-pdf-pypdfium2` | PDF rendering only | Lightweight PDF support |
+| `format-pdf-docling` | Advanced PDF parsing | Complex PDF layouts |
+
+#### Office Formats (office = docx + pptx + xlsx)
+
+| Extra | Description | Use Case |
+|-------|-------------|----------|
+| `format-office` | All Office formats | Microsoft Office documents |
+| `format-docx` | Microsoft Word documents | .docx files |
+| `format-pptx` | Microsoft PowerPoint | .pptx files |
+| `format-xlsx` | Microsoft Excel | .xlsx files |
+
+#### Web Formats (web = html + markdown)
+
+| Extra | Description | Use Case |
+|-------|-------------|----------|
+| `format-web` | HTML and Markdown | Web content |
+| `format-html` | HTML parsing | Web pages and HTML files |
+| `format-markdown` | Markdown parsing | .md files |
+
+#### Other Formats
+
+| Extra | Description | Use Case |
+|-------|-------------|----------|
+| `format-latex` | LaTeX documents | .tex files |
+| `format-xml-xbrl` | XBRL financial reports | Financial documents |
+| `format-html-render` | HTML rendering with Playwright | Dynamic web content |
+| `format-audio` | Audio transcription (Whisper) | .wav, .mp3 files |
+
+### OCR Engines
+
+| Extra | Description | Use Case |
+|-------|-------------|----------|
+| `feat-ocr-rapidocr` | RapidOCR (lightweight) | Fast OCR |
+| `feat-ocr-rapidocr-onnx` | RapidOCR with ONNX runtime | Optimized OCR |
+| `feat-ocr-easyocr` | EasyOCR | Multi-language OCR |
+| `feat-ocr-tesserocr` | Tesseract OCR | High-accuracy OCR |
+| `feat-ocr-mac` | macOS native OCR | macOS only |
+
+### Models
+
+| Extra | Description | Use Case |
+|-------|-------------|----------|
+| `models-local` | Local PyTorch models | GPU/CPU inference |
+| `models-remote` | Remote model serving (Triton) | Production deployments |
+| `models-onnxruntime` | ONNX Runtime acceleration | Optimized inference |
+| `models-vlm-inline` | Vision Language Models | Image understanding, inline processing |
+
+### Other features
+
+| Extra | Description | Use Case |
+|-------|-------------|----------|
+| `feat-chunking` | Document chunking | RAG applications |
+| `service-client` | Docling service client | Remote processing |
+
+
+## License
+
+MIT License - See [LICENSE](https://github.com/docling-project/docling/blob/main/LICENSE)
@@ -0,0 +1,7 @@
+# Docling
+
+This is a placeholder README for the `docling` meta-package.
+
+For the full README, see the [root README.md](../../README.md) in the repository.
+
+The actual README content is copied from the root during the build process.
@@ -0,0 +1,92 @@
+[build-system]
+requires = ["hatchling"]
+build-backend = "hatchling.build"
+
+[project]
+name = "docling"
+version = "2.91.0"  # DO NOT EDIT, updated automatically
+description = "SDK and CLI for parsing PDF, DOCX, HTML, and more, to a unified document representation for powering downstream workflows such as gen AI applications."
+license = "MIT"
+keywords = [
+  "docling",
+  "convert",
+  "document",
+  "pdf",
+  "docx",
+  "html",
+  "markdown",
+  "layout model",
+  "segmentation",
+  "table structure",
+  "table former",
+]
+classifiers = [
+  "Operating System :: MacOS :: MacOS X",
+  "Operating System :: POSIX :: Linux",
+  "Operating System :: Microsoft :: Windows",
+  "Development Status :: 5 - Production/Stable",
+  "Intended Audience :: Developers",
+  "Intended Audience :: Science/Research",
+  "Topic :: Scientific/Engineering :: Artificial Intelligence",
+  "Programming Language :: Python :: 3",
+  "Programming Language :: Python :: 3.10",
+  "Programming Language :: Python :: 3.11",
+  "Programming Language :: Python :: 3.12",
+  "Programming Language :: Python :: 3.13",
+  "Programming Language :: Python :: 3.14",
+]
+readme = "README.md"
+authors = [
+  { name = "Christoph Auer", email = "cau@zurich.ibm.com" },
+  { name = "Michele Dolfi", email = "dol@zurich.ibm.com" },
+  { name = "Maxim Lysak", email = "mly@zurich.ibm.com" },
+  { name = "Nikos Livathinos", email = "nli@zurich.ibm.com" },
+  { name = "Ahmed Nassar", email = "ahn@zurich.ibm.com" },
+  { name = "Panos Vagenas", email = "pva@zurich.ibm.com" },
+  { name = "Peter Staar", email = "taa@zurich.ibm.com" },
+]
+requires-python = '>=3.10,<4.0'
+
+# Meta-package: pulls in docling-slim with standard extras (includes CLI).
+# The `docling` Python module itself is provided by docling-slim.
+# CLI entry points are now defined in docling-slim's pyproject.toml.
+dependencies = [
+  'docling-slim[standard]==2.91.0',
+]
+
+[project.urls]
+homepage = "https://github.com/docling-project/docling"
+repository = "https://github.com/docling-project/docling"
+issues = "https://github.com/docling-project/docling/issues"
+changelog = "https://github.com/docling-project/docling/blob/main/CHANGELOG.md"
+
+[tool.uv.sources]
+# For local development: use workspace member
+docling-slim = { workspace = true }
+
+# Re-export slim extras for convenience
+[project.optional-dependencies]
+easyocr = ['docling-slim[feat-ocr-easyocr]==2.91.0']
+tesserocr = ['docling-slim[feat-ocr-tesserocr]==2.91.0']
+ocrmac = ['docling-slim[feat-ocr-mac]==2.91.0']
+vlm = ['docling-slim[models-vlm-inline]==2.91.0']
+rapidocr = ['docling-slim[feat-ocr-rapidocr-onnx]==2.91.0']
+chunking = ['docling-slim[feat-chunking]==2.91.0']
+format-audio = ['docling-slim[format-audio]==2.91.0']
+format-html-render = ['docling-slim[format-html-render]==2.91.0']
+models-remote = ['docling-slim[models-remote]==2.91.0']
+models-onnxruntime = ['docling-slim[models-onnxruntime]==2.91.0']
+format-xml-xbrl = ['docling-slim[format-xml-xbrl]==2.91.0']
+
+# Dependency-only wheel: no Python modules shipped here. All source lives in
+# the docling-slim wheel (built from the repo root). This avoids the prior
+# bug where both wheels shipped the same `docling/` module and collided on
+# install.
+[tool.hatch.build.targets.wheel]
+bypass-selection = true
+
+[tool.hatch.build.targets.sdist]
+only-include = ["pyproject.toml", "README.md"]
+
+[tool.uv]
+package = true
@@ -1,7 +1,11 @@
+[build-system]
+requires = ["hatchling"]
+build-backend = "hatchling.build"
+
 [project]
-name = "docling"
+name = "docling-slim"
 version = "2.91.0"  # DO NOT EDIT, updated automatically
-description = "SDK and CLI for parsing PDF, DOCX, HTML, and more, to a unified document representation for powering downstream workflows such as gen AI applications."
+description = "Modular version of the Docling package: SDK and CLI for parsing PDF, DOCX, HTML, and more, to a unified document representation for powering downstream workflows such as gen AI applications."
 license = "MIT"
 keywords = [
  "docling",
@@ -13,8 +17,6 @@ keywords = [
  "markdown",
  "layout model",
  "segmentation",
-  "table structure",
-  "table former",
 ]
 classifiers = [
  "Operating System :: MacOS :: MacOS X",
@@ -31,7 +33,7 @@ classifiers = [
  "Programming Language :: Python :: 3.13",
  "Programming Language :: Python :: 3.14",
 ]
-readme = "README.md"
+readme = "packages/docling-slim/README.md"
 authors = [
  { name = "Christoph Auer", email = "cau@zurich.ibm.com" },
  { name = "Michele Dolfi", email = "dol@zurich.ibm.com" },
@@ -42,40 +44,17 @@ authors = [
  { name = "Peter Staar", email = "taa@zurich.ibm.com" },
 ]
 requires-python = '>=3.10,<4.0'
+
+# MINIMAL BASE (8 packages) - ~50MB
 dependencies = [
-  'pydantic (>=2.0.0,<3.0.0)',
-  'docling-core[chunking] (>=2.73.0,<3.0.0)',
-  'docling-parse (>=5.3.2,<6.0.0)',
-  'docling-ibm-models>=3.13.0,<4',
-  'torch (>=2.2.2,<3.0.0)',
-  'torchvision (>=0,<1)',
-  'filetype (>=1.2.0,<2.0.0)',
-  'pypdfium2 (>=4.30.0,!=4.30.1,<6.0.0)',
-  'pydantic-settings (>=2.3.0,<3.0.0)',
-  'huggingface_hub (>=0.23,<2)',
-  'httpx (>=0.28,<1.0.0)',
-  'requests (>=2.32.2,<3.0.0)',
-  'ocrmac (>=1.0.0,<2.0.0) ; sys_platform == "darwin"',
-  'rapidocr (>=3.8,<4.0.0)',
-  'certifi (>=2024.7.4)',
-  'rtree (>=1.3.0,<2.0.0)',
-  'typer (>=0.12.5,<0.22.0)',
-  'python-docx (>=1.1.2,<2.0.0)',
-  'python-pptx (>=1.0.2,<2.0.0)',
-  'beautifulsoup4 (>=4.12.3,<5.0.0)',
-  'pandas (>=2.1.4,<4.0.0)',
-  'marko (>=2.1.2,<3.0.0)',
-  'openpyxl (>=3.1.5,<4.0.0)',
-  'lxml (>=4.0.0,<7.0.0)',
-  'pillow (>=10.0.0,<13.0.0)',
-  'tqdm (>=4.65.0,<5.0.0)',
-  'pluggy (>=1.0.0,<2.0.0)',
-  'pylatexenc (>=2.10,<3.0)',
-  'scipy (>=1.6.0,<2.0.0)',
-  "accelerate>=1.0.0,<2",
-  "polyfactory>=2.22.2",
-  "defusedxml (>=0.7.1, <0.8.0)",
-  "websockets (>=14.0,<17.0)",
+  'pydantic>=2.0.0,<3.0.0',
+  'docling-core>=2.73.0,<3.0.0',
+  'pydantic-settings>=2.3.0,<3.0.0',
+  'filetype>=1.2.0,<2.0.0',
+  'requests>=2.32.2,<3.0.0',
+  'certifi>=2024.7.4',
+  'pluggy>=1.0.0,<2.0.0',
+  'tqdm>=4.65.0,<5.0.0',
 ]

 [project.urls]
@@ -87,66 +66,207 @@ changelog = "https://github.com/docling-project/docling/blob/main/CHANGELOG.md"
 [project.entry-points.docling]
 "docling_defaults" = "docling.models.plugins.defaults"

+# CLI scripts (require cli extra: pip install docling-slim[cli])
 [project.scripts]
 docling = "docling.cli.main:app"
 docling-tools = "docling.cli.tools:app"

 [project.optional-dependencies]
-easyocr = ['easyocr (>=1.7,<2.0)']
-tesserocr = ['tesserocr (>=2.7.1,<3.0.0)']
-ocrmac = ['ocrmac (>=1.0.0,<2.0.0) ; sys_platform == "darwin"']
-htmlrender = ["playwright>=1.58.0"]
-vlm = [
-  'transformers (>=4.42.0,<6.0.0,!=5.0.*,!=5.1.*,!=5.2.*,!=5.3.*)',
-  'accelerate (>=1.2.1,<2.0.0)',
-  'mlx-vlm (>=0.4.3,<1.0.0) ; python_version >= "3.10" and sys_platform == "darwin" and platform_machine == "arm64"',
-  # 'vllm (>=0.10.0,<1.0.0) ; python_version >= "3.10" and python_version < "3.14" and sys_platform == "linux" and platform_machine == "x86_64"',
-  "qwen-vl-utils>=0.0.11",
-  "peft>=0.18.1",
+# ============================================================================
+# CORE COMPONENTS
+# ============================================================================
+convert-core = [
+  'numpy>=1.24.0,<3.0.0',
+  'pillow>=10.0.0,<13.0.0',
+  'rtree>=1.3.0,<2.0.0',
+  'scipy>=1.6.0,<2.0.0',
 ]
-rapidocr = [
-  'rapidocr (>=3.8,<4.0.0)',
-  'onnxruntime (>=1.7.0,<2.0.0) ; python_version < "3.14"',
+
+extract-core = [
+  'docling-slim[convert-core]',
+  'polyfactory>=2.22.2',
 ]
-onnxruntime = [
-  'onnxruntime (<1.24) ; python_version < "3.14" and sys_platform == "darwin"',
-  'onnxruntime-gpu (<1.24) ; python_version < "3.14" and (sys_platform == "linux" or sys_platform == "win32")',
+
+# ============================================================================
+# FORMAT SUPPORT
+# ============================================================================
+
+# --- PDF Formats ---
+format-pdf-pypdfium2 = [
+  'pypdfium2>=4.30.0,!=4.30.1,<6.0.0',
 ]
-asr = [
-    'mlx-whisper>=0.4.3 ; python_version >= "3.10" and sys_platform == "darwin" and platform_machine == "arm64"',
-    'openai-whisper>=20250625',
-    'numba>=0.63.0',  # Ensure Python 3.11+ compatibility (llvmlite constraint)
+
+format-pdf-docling = [
+  'pypdfium2>=4.30.0,!=4.30.1,<6.0.0',
+  'docling-parse>=5.3.2,<6.0.0',
 ]
-xbrl = [
-    "arelle-release (>=2.38.17,<3.0.0)",
+
+format-pdf = [
+  'docling-slim[format-pdf-pypdfium2,format-pdf-docling]',
 ]
-remote-serving = [
-    'tritonclient[grpc] (>=2.65.0,<3.0.0)',
+
+# --- Office Formats (office = docx + pptx + xlsx) ---
+format-docx = [
+  'python-docx>=1.1.2,<2.0.0',
+]
+
+format-pptx = [
+  'python-pptx>=1.0.2,<2.0.0',
+]
+
+format-xlsx = [
+  'openpyxl>=3.1.5,<4.0.0',
+]
+
+format-office = [
+  'docling-slim[format-docx,format-pptx,format-xlsx]',
+]
+
+# --- Web Formats (web = html + markdown) ---
+format-html = [
+  'beautifulsoup4>=4.12.3,<5.0.0',
+  'lxml>=4.0.0,<7.0.0',
+]
+
+format-markdown = [
+  'marko>=2.1.2,<3.0.0',
+]
+
+format-web = [
+  'docling-slim[format-html,format-markdown]',
+]
+
+# --- Other Formats ---
+format-latex = [
+  'pylatexenc>=2.10,<3.0',
+]
+
+format-xml-xbrl = [
+  'arelle-release>=2.38.17,<3.0.0',
+]
+
+format-html-render = [
+  'playwright>=1.58.0',
+]
+
+format-audio = [
+  'mlx-whisper>=0.4.3 ; python_version >= "3.10" and sys_platform == "darwin" and platform_machine == "arm64"',
+  'openai-whisper>=20250625',
+  'numba>=0.63.0',
+]
+
+# ============================================================================
+# OCR ENGINES (feat-ocr-*)
+# ============================================================================
+feat-ocr-rapidocr = [
+  'rapidocr>=3.8,<4.0.0',
+]
+
+feat-ocr-rapidocr-onnx = [
+  'rapidocr>=3.8,<4.0.0',
+  'onnxruntime>=1.7.0,<2.0.0 ; python_version < "3.14"',
+]
+
+feat-ocr-easyocr = [
+  'easyocr>=1.7,<2.0',
+  # easyocr declares scikit-image with no lower bound; without this pin,
+  # resolvers on Python 3.10 backtrack to 0.16.2 (2019), which has no
+  # Py3.10 wheels and fails to build from source.
+  'scikit-image>=0.19',
+]
+
+feat-ocr-tesserocr = [
+  'tesserocr>=2.7.1,<3.0.0',
+  'pandas>=2.1.4,<4.0.0',
+]
+
+feat-ocr-mac = [
+  'ocrmac>=1.0.0,<2.0.0 ; sys_platform == "darwin"',
+]
+
+# ============================================================================
+# MODELS
+# ============================================================================
+models-local = [
+  'torch>=2.2.2,<3.0.0',
+  'torchvision>=0,<1',
+  'docling-ibm-models>=3.13.0,<4',
+  'accelerate>=1.0.0,<2',
+  'huggingface_hub>=0.23,<2',
+  'defusedxml>=0.7.1,<0.8.0',
+]
+
+models-remote = [
+  'tritonclient[grpc]>=2.65.0,<3.0.0',
+]
+
+models-onnxruntime = [
+  'onnxruntime<1.24 ; python_version < "3.14" and sys_platform == "darwin"',
+  'onnxruntime-gpu<1.24 ; python_version < "3.14" and (sys_platform == "linux" or sys_platform == "win32")',
+]
+
+# Vision Language Models for inline processing
+models-vlm-inline = [
+  'transformers>=4.42.0,<6.0.0,!=5.0.*,!=5.1.*,!=5.2.*,!=5.3.*',
+  'accelerate>=1.2.1,<2.0.0',
+  'mlx-vlm>=0.4.3,<1.0.0 ; python_version >= "3.10" and sys_platform == "darwin" and platform_machine == "arm64"',
+  'qwen-vl-utils>=0.0.11',
+  'peft>=0.18.1',
+]
+
+# ============================================================================
+# FEATURES
+# ============================================================================
+# Document chunking for RAG applications
+feat-chunking = [
+  'docling-core[chunking]>=2.73.0,<3.0.0',
+]
+
+service-client = [
+  'httpx>=0.28,<1.0.0',
+  'websockets>=14.0,<17.0',
+]
+
+cli = [
+  'typer>=0.12.5,<0.22.0',
+  'rich>=13.0.0',
+]
+
+# ============================================================================
+# CONVENIENCE BUNDLES
+# ============================================================================
+standard = [
+  'docling-slim[format-pdf,models-local,feat-ocr-rapidocr,format-office,format-web,format-latex,feat-chunking,extract-core,service-client,cli]',
+]
+
+all = [
+  'docling-slim[standard,models-vlm-inline,format-audio,format-html-render,format-xml-xbrl,models-remote,models-onnxruntime,feat-ocr-easyocr,feat-ocr-tesserocr,feat-ocr-mac]',
 ]

 [dependency-groups]
 dev = [
-    "pre-commit~=3.7",
-    "mypy~=1.10",
-    "types-setuptools~=70.3",
-    "pandas-stubs~=2.1",
-    "types-openpyxl~=3.1",
-    "types-requests~=2.31",
-    "boto3-stubs~=1.37",
-    "types-urllib3~=1.26",
-    "types-tqdm~=4.67",
-    "coverage~=7.6",
-    "pytest~=8.3",
-    "pytest-cov>=6.1.1",
-    "pytest-dependency~=0.6",
-    "pytest-durations~=1.6.1",
-    "pytest-xdist~=3.3",
-    "ipykernel~=6.29",
-    "ipywidgets~=8.1",
-    "nbqa~=1.9",
-    "python-semantic-release~=7.32",
-    "types-defusedxml (>=0.7.0.20250822, <0.8.0)",
+  "pre-commit~=3.7",
+  "mypy~=1.10",
+  "types-setuptools~=70.3",
+  "pandas-stubs~=2.1",
+  "types-openpyxl~=3.1",
+  "types-requests~=2.31",
+  "boto3-stubs~=1.37",
+  "types-urllib3~=1.26",
+  "types-tqdm~=4.67",
+  "coverage~=7.6",
+  "pytest~=8.3",
+  "pytest-cov>=6.1.1",
+  "pytest-dependency~=0.6",
+  "pytest-durations~=1.6.1",
+  "pytest-xdist~=3.3",
+  "ipykernel~=6.29",
+  "ipywidgets~=8.1",
+  "nbqa~=1.9",
+  "python-semantic-release~=7.32",
+  "types-defusedxml>=0.7.0.20250822,<0.8.0",
 ]
+
 docs = [
  "mkdocs-material~=9.5",
  "mkdocs-jupyter>=0.25,<0.26",
@@ -155,6 +275,7 @@ docs = [
  "mkdocstrings[python]~=0.27",
  "griffe-pydantic~=1.1",
 ]
+
 examples = [
  "datasets~=2.21",
  "python-dotenv~=1.0",
@@ -162,53 +283,52 @@ examples = [
  "langchain-milvus~=0.1",
  "langchain-text-splitters>=0.2",
  "modelscope>=1.29.0",
-  'gliner>=0.2.21 ; python_version < "3.14"',  # gliner depends on onnxruntime which is not available on py3.14
-]
-constraints = [
-  'numba >=0.63.0',
-  'langchain-core >=0.3.81',
-  'pandas (>=2.1.4,<3.0.0); python_version < "3.11"',
-  'pandas (>=2.1.4,<4.0.0); python_version >= "3.11"',
+  'gliner>=0.2.21 ; python_version < "3.14"',
 ]

+constraints = [
+  'numba>=0.63.0',
+  'langchain-core>=0.3.81',
+  'pandas>=2.1.4,<3.0.0 ; python_version < "3.11"',
+  'pandas>=2.1.4,<4.0.0 ; python_version >= "3.11"',
+]
+
+[tool.uv.workspace]
+members = ["packages/docling"]
+
+[tool.uv.sources]
+docling = { workspace = true }

 [tool.uv]
 package = true
 default-groups = "all"

-[tool.setuptools.packages.find]
-include = ["docling*"]
+[tool.hatch.build.targets.wheel]
+packages = ["docling"]
+
+[tool.hatch.build.targets.sdist]
+only-include = ["docling", "pyproject.toml", "README.md", "LICENSE"]

 [tool.ruff]
 target-version = "py310"
 line-length = 88
 respect-gitignore = true

-# extend-exclude = [
-#     "tests",
-# ]
-
 [tool.ruff.format]
 skip-magic-trailing-comma = false

 [tool.ruff.lint]
 select = [
-  # "B",  # flake8-bugbear
  "C",  # flake8-comprehensions
  "C9", # mccabe
-  # "D",  # flake8-docstrings
  "E",   # pycodestyle errors (default)
  "F",   # pyflakes (default)
  "I",   # isort
  "PD",  # pandas-vet
  "PIE", # pie
-  # "PTH", # pathlib
  "Q", # flake8-quotes
-  # "RET", # return
  "RUF", # Enable all ruff-specific checks
-  # "SIM", # simplify
  "S307", # eval
-  # "T20",  # (disallow print statements) keep debugging statements out of the codebase
  "W",     # pycodestyle warnings
  "ASYNC", # async
  "UP",    # pyupgrade
@@ -227,38 +347,23 @@ ignore = [
  "UP035",  # `typing.Set` is deprecated, use `set` instead"
 ]

-#extend-select = []
-
 [tool.ruff.lint.pep8-naming]
 classmethod-decorators = [
-  # Allow Pydantic's `@validator` decorator to trigger class method treatment.
  "pydantic.validator",
 ]

 [tool.ruff.lint.per-file-ignores]
 "__init__.py" = ["E402", "F401"]
-"tests/*.py" = ["ASYNC"]         # Disable ASYNC check for tests
+"tests/*.py" = ["ASYNC"]

 [tool.ruff.lint.mccabe]
 max-complexity = 30

-# [tool.ruff.lint.isort.sections]
-# "docling" = ["docling_core", "docling_ibm_models", "docling_parse"]
-
 [tool.ruff.lint.isort]
 combine-as-imports = true
-# section-order = [
-#   "future",
-#   "standard-library",
-#   "third-party",
-#   "docling",
-#   "first-party",
-#   "local-folder",
-# ]

 [tool.mypy]
 pretty = true
-# strict = true
 no_implicit_optional = true
 plugins = "pydantic.mypy"
 python_version = "3.10"