### Quick start for Docling CLI and evaluation

Source: https://github.com/docling-project/docling/blob/main/docs/examples/agent_skill/docling-document-intelligence/README.md

Install dependencies and perform document conversion followed by quality evaluation. Use the --pipeline vlm flag for vision-model pipelines.

```bash
pip install docling docling-core

# Convert to Markdown
docling https://arxiv.org/pdf/2408.09869 --output /tmp/

# Convert to JSON
docling https://arxiv.org/pdf/2408.09869 --to json --output /tmp/

# Evaluate quality
python3 scripts/docling-evaluate.py /tmp/2408.09869.json --markdown /tmp/2408.09869.md
```

--------------------------------

### Start vLLM Server on Linux

Source: https://github.com/docling-project/docling/blob/main/docs/getting_started/rtx.md

Use this command to start the vLLM inference server with optimized parameters for maximum performance on Linux systems. Ensure vLLM is installed and the model is accessible.

```bash
vllm serve ibm-granite/granite-docling-258M \
  --host 127.0.0.1 --port 8000 \
  --max-num-seqs 512 \
  --max-num-batched-tokens 8192 \
  --enable-chunked-prefill \
  --gpu-memory-utilization 0.9
```

--------------------------------

### Install the Full Docling Package

Source: https://github.com/docling-project/docling/blob/main/packages/docling-slim/README.md

Install the recommended, full-featured Docling package with all standard capabilities.

```bash
pip install docling
```

--------------------------------

### Install Docling

Source: https://github.com/docling-project/docling/blob/main/docs/examples/service_client/docling_service_client_file_quickstart.ipynb

Installs the docling library. Run this cell before proceeding.

```python
%pip install -qU docling
```

--------------------------------

### Install Docling Slim Service Client

Source: https://github.com/docling-project/docling/blob/main/packages/docling-slim/README.md

Install Docling Slim with the service client for interacting with the Docling Serve API.

```bash
pip install docling-slim[service-client]
```

--------------------------------

### Install Development Dependencies

Source: https://github.com/docling-project/docling/blob/main/docs/getting_started/installation.md

Install all necessary dependencies for development, including optional extras, using the `uv` package manager.

```bash
uv sync --all-extras
```

--------------------------------

### Install Docling ASR Pipeline

Source: https://github.com/docling-project/docling/blob/main/docs/usage/processing_audio_media.md

Install the ASR pipeline alongside the base Docling package using pip or uv.

```bash
pip install "docling[asr]"
```

```bash
uv add "docling[asr]"
```

--------------------------------

### Install Docling with XBRL Support

Source: https://github.com/docling-project/docling/blob/main/docs/examples/xbrl_conversion.ipynb

Installs the Docling package with XBRL parsing capabilities using pip. This is the initial setup step required before working with XBRL documents.

```python
%pip install -q docling
```

--------------------------------

### Configure DocumentConverter with Tesseract OCR

Source: https://github.com/docling-project/docling/blob/main/docs/getting_started/installation.md

Example of setting up DocumentConverter to use Tesseract OCR for PDF processing. Ensure Tesseract is installed and configured.

```python
from docling.datamodel.base_models import InputFormat
from docling.datamodel.pipeline_options import (
    TesseractOcrOptions,
    PdfPipelineOptions,
)
from docling.document_converter import DocumentConverter, PdfFormatOption

pipeline_options = PdfPipelineOptions()
pipeline_options.do_ocr = True
pipeline_options.ocr_options = TesseractOcrOptions()  # Use Tesseract

doc_converter = DocumentConverter(
    format_options={InputFormat.PDF: PdfFormatOption(pipeline_options=pipeline_options)}
)
```

--------------------------------

### Install and Serve vLLM Backend

Source: https://github.com/docling-project/docling/blob/main/docs/examples/agent_skill/docling-document-intelligence/pipelines.md

Sets up the vLLM server to host the Granite Docling model for high-throughput batch processing.

```bash
pip install vllm
vllm serve ibm-granite/granite-docling-258M
```

--------------------------------

### Configure OpenAITokenizer for Chunking

Source: https://github.com/docling-project/docling/blob/main/docs/examples/agent_skill/docling-document-intelligence/SKILL.md

This example shows how to set up OpenAITokenizer using tiktoken for compatibility with OpenAI embedding models, noting the required installation.

```python
import tiktoken
from docling_core.transforms.chunker.tokenizer.openai import OpenAITokenizer

tokenizer = OpenAITokenizer(
    tokenizer=tiktoken.encoding_for_model("text-embedding-3-small"),
    max_tokens=8192,
)
# Requires: pip install 'docling-core[chunking-openai]'
```

--------------------------------

### Run MkDocs Documentation Server

Source: https://github.com/docling-project/docling/blob/main/CONTRIBUTING.md

Start the local documentation server, accessible at `http://localhost:8000`.

```bash
mkdocs serve
```

--------------------------------

### Install Docling Agent Skill

Source: https://github.com/docling-project/docling/blob/main/docs/examples/agent_skill/docling-document-intelligence/EXAMPLE.md

Copy the Docling agent skill directory into your AI agent's skills folder to enable its functionality. This setup allows agents like Cursor or Claude to load the skill.

```bash
cp -r docs/examples/agent_skill/docling-document-intelligence ~/.cursor/skills/

# Or copy from another machine / archive into e.g. ~/.claude/skills/
```

--------------------------------

### Load and Display Example Document Image

Source: https://github.com/docling-project/docling/blob/main/docs/examples/extraction.ipynb

Define a URL for an example image and use `IPython.display.HTML` to render it within the notebook for inspection.

```python
file_path = (
    "https://upload.wikimedia.org/wikipedia/commons/9/9f/Swiss_QR-Bill_example.jpg"
)
display.HTML(f"<img src='{file_path}' height='1000'>")
```

--------------------------------

### Install and Run docling-serve (Local Engine)

Source: https://github.com/docling-project/docling/blob/main/docs/usage/api_server/deployment.md

Installs the necessary package and runs the docling-serve API server using the default local compute engine. The `dev` command is suitable for development with auto-reloading.

```sh
pip install "docling-serve[ui]"
docling-serve run --enable-ui      # production-style: reload off, binds 0.0.0.0, UI off by default
# docling-serve dev                # dev: auto-reload, binds 127.0.0.1, UI on (localhost only)
```

--------------------------------

### Install LlamaIndex Docling Integration Dependencies

Source: https://github.com/docling-project/docling/blob/main/docs/examples/rag_llamaindex.ipynb

Install required Python packages for the RAG pipeline including LlamaIndex core, Docling readers/parsers, Hugging Face embeddings and LLM support, Milvus vector store, and file readers. The --no-warn-conflicts flag is specific to Colab environments to avoid conflicts with pre-installed packages.

```python
%pip install -q --progress-bar off --no-warn-conflicts llama-index-core llama-index-readers-docling llama-index-node-parser-docling llama-index-embeddings-huggingface llama-index-llms-huggingface-api llama-index-vector-stores-milvus llama-index-readers-file python-dotenv
```

--------------------------------

### Define Plugin Entrypoint in setup.py

Source: https://github.com/docling-project/docling/blob/main/docs/concepts/plugins.md

Configure plugin entrypoint using setup.py format with setuptools. This registers your plugin with the Docling ecosystem by specifying entry_points in the setup() function call.

```python
from setuptools import setup

setup(
    # ...,
    entry_points = {
        'docling': [
            'your_plugin_name = "your_package.module"'
        ]
    }
)
```

--------------------------------

### Setup Experiment with Temporary Folder and Articles

Source: https://github.com/docling-project/docling/blob/main/docs/examples/dpk-ingest-chunk-tokenize.ipynb

Creates a temporary directory for storing downloaded articles and initializes the list of articles to retrieve from Wikimedia. Calls load_corpus to download the specified articles and asserts that at least one document was successfully downloaded.

```python
import os
import tempfile

datafolder = tempfile.mkdtemp(dir=os.getcwd())
articles = ["Science,_technology,_engineering,_and_mathematics"]
assert load_corpus(articles, datafolder) > 0, "Faild to download any documents"
```

--------------------------------

### Build LlamaIndex RAG Pipeline with DoclingReader (JSON) and Milvus for Richer Grounding

Source: https://github.com/docling-project/docling/blob/main/docs/examples/rag_llamaindex.ipynb

This snippet demonstrates an advanced RAG pipeline leveraging Docling's rich native JSON format. It configures the DoclingReader to export as JSON and uses a specialized DoclingNodeParser to process this format, enabling more granular document-level grounding in the query results. The setup for the vector store, indexing, and querying remains similar to the Markdown example.

```python
from llama_index.node_parser.docling import DoclingNodeParser

reader = DoclingReader(export_type=DoclingReader.ExportType.JSON)
node_parser = DoclingNodeParser()

vector_store = MilvusVectorStore(
    uri=str(Path(mkdtemp()) / "docling.db"),  # or set as needed
    dim=embed_dim,
    overwrite=True,
)
index = VectorStoreIndex.from_documents(
    documents=reader.load_data(SOURCE),
    transformations=[node_parser],
    storage_context=StorageContext.from_defaults(vector_store=vector_store),
    embed_model=EMBED_MODEL,
)
result = index.as_query_engine(llm=GEN_MODEL).query(QUERY)
print(f"Q: {QUERY}\nA: {result.response.strip()}\n\nSources:")
```

--------------------------------

### Configure Flash Attention 2 for Docling Models

Source: https://github.com/docling-project/docling/blob/main/docs/faq/index.md

These examples show how to enable Flash Attention 2 when running Docling models on CUDA devices. You can activate it by setting an environment variable before execution or by programmatically configuring `VlmPipelineOptions` in Python. This feature requires the `flash-attn` package to be installed and can significantly enhance model performance.

```shell
DOCLING_CUDA_USE_FLASH_ATTENTION2=1
```

```python
from docling.datamodel.accelerator_options import (
    AcceleratorOptions,
)

pipeline_options = VlmPipelineOptions(
    accelerator_options=AcceleratorOptions(cuda_use_flash_attention2=True)
)
```

--------------------------------

### Install Docling with uv

Source: https://github.com/docling-project/docling/blob/main/docs/getting_started/installation.md

Use this command to add the Docling package to your project using uv.

```bash
uv add docling
```

--------------------------------

### Install Tesserocr with specific options

Source: https://github.com/docling-project/docling/blob/main/docs/getting_started/installation.md

Provides alternative pip installation commands for the Tesserocr package to resolve potential installation issues. It forces a source build.

```bash
pip uninstall tesserocr
pip install --no-binary :all: tesserocr
```

--------------------------------

### Install VLM Dependencies for Docling

Source: https://github.com/docling-project/docling/blob/main/docs/examples/agent_skill/docling-document-intelligence/pipelines.md

Installs the necessary packages for local VLM inference using PyTorch and Transformers. You can use the [vlm] extra or install components manually.

```bash
pip install docling[vlm]
# or manually:
pip install torch transformers accelerate
```

--------------------------------

### Install Docling Slim with PDF and Local Models

Source: https://github.com/docling-project/docling/blob/main/packages/docling-slim/README.md

Install Docling Slim with support for PDF parsing and local PyTorch models.

```bash
pip install docling-slim[format-pdf,models-local]
```