Install Docling on macOS: Apple Silicon, MPS, MLX & Intel Macs

Install Docling on a Mac — Apple Silicon does the heavy lifting

On Apple Silicon, a plain pip install docling already includes GPU acceleration through PyTorch's MPS backend, and two Mac-only extras — Apple Vision OCR and MLX models — make it even faster. Two version details are worth knowing first.

Verified on Docling 2.148.0Updated 6 Nov 20265 min read

In short

pip install docling in a Python 3.10–3.14 venv; MPS is used automatically (--device auto).
PyTorch wheels newer than 2.11 need macOS 14 or later on Apple Silicon; on macOS 11–13 pip installs torch 2.11.0.
Intel Macs resolve torch 2.2.2, the last Intel macOS build — Docling still installs, but expect slower, older PyTorch.
Add docling[ocrmac] for Apple Vision OCR and docling[vlm] for MLX-accelerated Granite-Docling.

Install#

bash

brew install python@3.12                 # or the python.org installer
python3.12 -m venv ~/docling-env && source ~/docling-env/bin/activate
pip install -U pip
pip install "docling[ocrmac]"            # Apple Vision OCR included
docling --version

No separate PyTorch step is needed on macOS: PyPI's macOS wheels do not pull NVIDIA libraries, and they include MPS support.

Which PyTorch you will get#

Mac	PyTorch resolved for Docling 2.148.0	Why
Apple Silicon, macOS 14+	2.15.0 (latest)	Wheels from 2.12.0 on are tagged `macosx_14_0_arm64`
Apple Silicon, macOS 11–13	2.11.0	Last version with `macosx_11_0_arm64` wheels
Intel (x86_64)	2.2.2	Last PyTorch release with Intel macOS wheels

Resolutions checked with uv's resolver for Python 3.12 and confirmed against the wheel tags on PyPI. Docling requires torch>=2.2.2, which is why Intel Macs still work.

Apple Silicon acceleration#

Layout, tables, OCR models: run on MPS automatically. Force with --device mps.
Apple Vision OCR: --ocr-engine ocrmac; languages use Apple's codes such as en-US, zh-Hans, ja-JP, or iso: tags. --ocr-option recognition=fast trades accuracy for speed.
MLX for VLMs: pip install "docling[vlm]" adds mlx-vlm on Apple Silicon; presets with an MLX build (Granite-Docling, SmolDocling, Qwen, Pixtral, Gemma…) use it automatically.
MLX for audio: docling[asr] adds mlx-whisper; pick a whisper_*_mlx model with --asr-model.

bash

docling convert scan.pdf --ocr-engine ocrmac --ocr-lang ja-JP
docling convert report.pdf --pipeline vlm --vlm-model granite_docling
docling convert meeting.m4a --pipeline asr --asr-model whisper_turbo_mlx

Optional tools via Homebrew#

bash

brew install tesseract tesseract-lang     # Tesseract engine + all languages
brew install --cask libreoffice           # .doc, .ppt, .xls, .rtf
brew install ffmpeg                       # audio and video

Frequently asked questions

Does Docling use the Apple Silicon GPU?

Yes. PyTorch's MPS backend ships in the standard macOS wheels, and Docling's default --device auto picks MPS when CUDA is not available. You can force it with --device mps.

What is the best OCR engine for Docling on a Mac?

ocrmac, which uses Apple's built-in Vision framework: no model download and good multilingual support. Install docling[ocrmac] and pass --ocr-engine ocrmac. On macOS the automatic engine also tries ocrmac first when it is installed.

Can I run Docling on an Intel Mac?

Yes. Docling 2.148.0 installs, but PyTorch stopped publishing Intel macOS wheels after 2.2.2, so pip installs that version. Everything runs on the CPU.

Sources

Was this accurate? We re-verify pages when Docling releases change behaviour. If a command fails for you, email hello@docling.org with your Docling version and the error — corrections are usually live within a day.

Install Docling on a Mac — Apple Silicon does the heavy lifting

Install#

Which PyTorch you will get#

Apple Silicon acceleration#

Optional tools via Homebrew#

Frequently asked questions

Sources

Keep going

Install

VLM & Granite-Docling

OCR

Performance

Troubleshooting

Docker