Docling Command Finder

Flags used

• --to md Export Markdown (the default output format).

Expected output

# Annual Report

## Revenue

| Year | Revenue |
|------|--------:|
| 2025 | $12M |
| 2026 | $15M |

Variations

Flags used

• --to md Markdown output.

Expected output

## Docling Technical Report

The conversion pipeline analyses layout, reading order and tables…

Variations

Flags used

• --to json Lossless JSON representation of the document tree.

Expected output

{
  "schema_name": "DoclingDocument",
  "texts": [ ... ],
  "tables": [ ... ],
  "pictures": [ ... ]
}

Variations

Flags used

• --version Shows Docling, docling-core and docling-ibm-models versions.

Expected output

Docling version: 2.129.0
Docling Core version: 2.x.x
Docling IBM Models version: 3.x.x
Python: cpython-312 …

Variations

Expected output

Usage: docling convert [OPTIONS] SOURCE

  --from TEXT        Input formats to accept…
  --to TEXT          Output formats…
  --ocr-engine TEXT  The OCR engine to use…

Variations

Flags used

• --output Directory where results are written (default: current directory).

Expected output

./out/report.md

Variations

Flags used

• --to Repeatable. Available: md, json, yaml, html, html_split_page, text, doctags, vtt, doclang, dclx, chunks, latex.

Expected output

report.md  report.json  report.html

Variations

Flags used

• --output Folder for the converted results.
• --abort-on-error Optional: stop at the first file that fails.

Expected output

Converting ./inbox/a.pdf … done
Converting ./inbox/b.docx … done

Variations

Flags used

• source Accepts one or more local paths, directories or URLs.

Expected output

a.md  b.md  c.md  written to ./out

Variations

Flags used

• --all Download every available model (large).

Expected output

Downloading layout model…
Downloading tableformer model…
Models cached in $HOME/.cache/docling/models

Variations

Flags used

• --to md Markdown output.

Expected output

# Service Agreement

1. Scope
2. Payment terms…

Variations

Flags used

• --page-range Optional: convert only e.g. the first slides (1-5).

Expected output

## Slide 1 — Overview

Bullet one
Bullet two

Variations

Flags used

• --page-range Optional: limit which sheets are converted.

Expected output

## Sheet 1

| Region | Q1 | Q2 |
|--------|----|----|
| EMEA   | 12 | 15 |

Variations

Flags used

• --html-image-fetch Fetch images referenced by HTML/EPUB (none, local, remote, all).

Expected output

# Page title

Body text converted from HTML…

Variations

Flags used

• --to md Markdown table output.

Expected output

| name | score |
|------|------:|
| Ada  | 98    |

Variations

Flags used

• --html-image-fetch Fetch images embedded in the EPUB.

Expected output

# Chapter 1

Long-form text…

Variations

Flags used

• --to html HTML output.

Expected output

<h1>Notes</h1>
<p>…</p>

Variations

Flags used

• --tikz-engine Set to 'tectonic' to rasterize tikzpicture diagrams.

Expected output

# Introduction

The math is preserved as LaTeX where possible…

Variations

Flags used

• --ocr-mode full_page OCR the whole image.
• --ocr-engine Optionally pick a specific engine.

Expected output

Text recognised from the image…

Variations

Flags used

• --ocr-mode full_page OCR every page from its rendered image and replace detected text.

Expected output

Text reconstructed from the scanned page images…

Variations

Flags used

• --ocr-engine auto, easyocr, rapidocr, tesserocr, tesseract, ocrmac, nemotron-ocr, kserve_v2_ocr.

Expected output

Using OCR engine: rapidocr

Variations

Flags used

• --ocr-lang Comma-separated native engine codes (e.g. deu,fra) or BCP-47 tags prefixed with iso: (e.g. iso:zh-Hans).

Expected output

Using OCR languages: deu, fra

Variations

Flags used

• --no-ocr Turn OCR off; the embedded text layer is used as-is.

Expected output

Skipping OCR (digital text layer detected)…

Variations

Flags used

• --ocr-mode layout_regions Feed detected layout regions to the OCR engine.

Expected output

OCR applied to detected layout regions…

Variations

Flags used

• --psm Page Segmentation Mode (0-13). Applies to Tesseract engines.

Expected output

Tesseract PSM 6 — assume a single uniform block of text.

Variations

Flags used

• --table-mode accurate (TableFormer, default) or fast.

Expected output

Rough table grid, produced faster…

Variations

Flags used

• --no-tables Do not run the table structure model.

Expected output

Tables rendered as plain text flow…

Variations

Flags used

• --table-structure-engine docling_tableformer (default), docling_tableformer_v2, granite_vision_table.

Expected output

Using table structure engine: docling_tableformer_v2

Variations

Flags used

• --enrich-code Detect and label code blocks.
• --enrich-formula Extract formulas as LaTeX.

Expected output

```python
def hello(): …
```

$$ E = mc^2 $$

Variations

Flags used

• --enrich-picture-description Run a picture-description model over detected pictures.

Expected output

<!-- picture: a bar chart showing revenue growth from 2020 to 2026 -->

Variations

Flags used

• --enrich-picture-classes Run the picture classification model.

Expected output

<!-- picture class: chart -->

Variations

Flags used

• --enrich-chart-extraction Extract chart data from bar, pie and line charts.

Expected output

<!-- chart: category | value -->
<!-- 2025 | 12 -->

Variations

Flags used

• --to chunks Export RAG-ready chunks.
• --chunks-type hybrid (default) or hierarchical.

Expected output

{ "text": "…", "meta": { "headings": ["Revenue"] } }

Variations

Flags used

• --chunks-max-tokens Max tokens per chunk (defaults to the tokenizer's limit).
• --chunks-tokenizer HuggingFace tokenizer (default sentence-transformers/all-MiniLM-L6-v2).

Expected output

Chunks sized to the embedding model's token limit…

Variations

Flags used

• --pipeline vlm Select the VLM pipeline.
• --vlm-model Preset: granite_docling (default), smoldocling, deepseek_ocr, granite_vision, and more.

Expected output

Markdown generated page-by-page by the vision model…

Variations

Flags used

• --vlm-max-new-tokens Override max_new_tokens for VLM generation.

Expected output

Long, dense pages no longer get cut off…

Variations

Flags used

• --pipeline asr Select the speech-recognition pipeline.
• --asr-model Whisper size: whisper_tiny (default) … whisper_large.

Expected output

00:00:00 — Welcome to the show…

Variations

Flags used

• --to vtt WebVTT subtitle output with timestamps.

Expected output

WEBVTT

00:00:00.000 --> 00:00:04.000
Hello and welcome…

Variations

Flags used

• --video-sampling-mode fixed (default) or scene.
• --video-frame-interval Seconds between frames in fixed mode (default 10).

Expected output

Frames sampled at scene changes…

Variations

Flags used

• --video-diarization Enable speaker diarization; requires resemblyzer.

Expected output

[SPEAKER_00] …
[SPEAKER_01] …

Variations

Flags used

• --image-export-mode embedded (base64, default), placeholder, or referenced (PNG files).

Expected output

./out/report.md + ./out/report_artifacts/*.png

Variations

Flags used

• --to doctags DocTags markup output.

Expected output

<doctag><page_1><section_header_level_1>Annual Report</section_header_level_1>…

Variations

Flags used

• --to html_split_page One HTML file per page.

Expected output

./out/report_1.html  report_2.html …

Variations

Flags used

• --show-layout Show item bounding boxes on page images.

Expected output

Page images with coloured layout boxes…

Variations

Flags used

• --device cuda Use the NVIDIA GPU.
• --num-threads CPUs used for model inference (default 4).

Expected output

Using accelerator device: cuda

Variations

Flags used

• --device mps Use Apple Metal Performance Shaders.

Expected output

Using accelerator device: mps

Variations

Flags used

• --page-batch-size Pages processed in one batch (default 4).

Expected output

Processing 16 pages per batch…

Variations

Flags used

• --document-timeout Timeout per document, in seconds.

Expected output

Timed out after 120s — moving to the next file…

Variations

Flags used

• --profiling Summarise profiling details for all stages.
• --save-profiling Save profiling summaries to JSON.

Expected output

layout: 3.2s  ocr: 1.1s  tableformer: 0.9s  total: 5.4s

Variations

Flags used

• --page-range e.g. 1-4 (page numbers start at 1). Honoured by PDF, XLSX and PPTX.

Expected output

Converting pages 1-4 only…

Variations

Flags used

• --pdf-password Password for protected PDF documents.

Expected output

Decrypting and converting locked.pdf…

Variations

Flags used

• --pdf-backend docling_parse (default) or pypdfium2.

Expected output

Using PDF backend: pypdfium2

Variations

Flags used

• --artifacts-path Location of pre-downloaded model artifacts.

Expected output

Loading models from /opt/docling/models…

Variations

Flags used

• DOCLING_ARTIFACTS_PATH Directory holding the pre-downloaded models.
• HF_HUB_OFFLINE Stop HuggingFace downloads and use the local cache only.

Expected output

Conversion completes with no outbound requests…

Variations

Flags used

• --enable-ui Serve the built-in web UI alongside the API.

Expected output

Uvicorn running on http://0.0.0.0:5001  (docs at /docs)

Variations

Flags used

• --service-url Base URL of docling-serve (or DOCLING_SERVICE_URL).
• --api-key Optional API key (or DOCLING_SERVICE_API_KEY).

Expected output

submitting job… polling… report.md written

Variations

Flags used

• --from=docling-mcp Runs the MCP server package without a global install.

Expected output

docling-mcp server ready (stdio)

Variations

Flags used

• -v / --verbose Repeat for more detail: -v info, -vv debug.
• -q / --quiet Silence per-file progress (warnings and errors remain).

Expected output

DEBUG docling.pipeline… loading layout model

Variations

Flags used

• --debug-visualize-layout Visualise layout clusters.
• --debug-visualize-tables Visualise table cells.
• --debug-visualize-ocr Visualise OCR cells.
• --debug-visualize-cells Visualise PDF cells.

Expected output

Annotated page images written next to the output…

Variations