MinerU Document Extraction Tools

MinerU Document Extraction Tools

Extracts content from PDFs, Office files and other documents with a hosted MinerU demo that runs on ZeroGPU. For developers and data teams who turn documents into structured text.

Made by opendatalab on Hugging Face. These notes are Mindket's, written from its Space page and README; the app's own files and license are on Hugging Face.

About

What it is: MinerU Document Extraction Tools is a Gradio demo of MinerU, a document extraction tool, running on Hugging Face's ZeroGPU hardware. You upload a document and the app pulls its content out, returning results as Middle JSON together with the assets it saves. You can try it in a browser with nothing to install.

Who it suits: It suits developers, data teams and researchers who need the text and structure of documents in a form a program can read.

Industry | Who uses it | How they use it | Example
Software | Developers | Test document extraction before building it into an app | Upload a short PDF and read the Middle JSON
Data and analytics | Data teams | Turn reports and files into structured content | Process an Office file and look at the result
AI research | Researchers | Check how an extraction pipeline handles different inputs | Compare a PDF run with an OCR run

If you only need to read a document, you can skip it, since the output is meant for programs.

Where to use it:

• A web browser, on the Space's page on Hugging Face
• A local copy of the files, run on Python 3.12 under Windows, macOS or Linux

Use case | Who | What they get back
Extract a PDF | Developers | Structured content as Middle JSON plus saved assets
Read scanned pages | Data teams | Text from pages through OCR
Process office files | Analysts | Content from native Office, HTML or OFD input

How to set it up:

• A web browser, which is all the online page needs
• Python 3.12 and the packages in requirements.txt, for a local copy
• Time at first start, because the launcher builds an isolated MinerU environment and keeps a cache in your home folder, which MINERU_RUNTIME_CACHE can move

1. Open the Space's page on Hugging Face to use it online. Nothing needs installing.
2. For a local copy, install the Hugging Face hub package and download the Space's files.
3. Move into the downloaded folder.
4. Install Gradio 6.28.0, then the packages from requirements.txt.
5. Start the app with python app.py.

pip install -U huggingface_hub
hf download opendatalab/MinerU --repo-type space --local-dir MinerU
cd MinerU
pip install gradio==6.28.0
pip install -r requirements.txt
python app.py

How to use it: Open the page, upload a document and start the extraction. The top level, Basic, caps a request at 20 pages and stops a GPU parse after 60 seconds. The page shows queue and processing status while it works.

Tool | What it does
Flash TXT | Handles plain text input on the CPU
Native Office, HTML and OFD | Handles those file types on the CPU through a PDFium sidecar
Flash OCR | Reads text from page images on ZeroGPU
Basic PDF | Parses PDF requests on ZeroGPU, up to 20 pages

Developers can run the project's tests in an environment that has MinerU and pytest installed.

python -m pytest tests

Example: Upload a short PDF and choose the Basic level. The app parses the pages on a GPU and returns the structured result as Middle JSON, with the assets it saved from the document.

For newcomers and experts: Open the Space page and try a short PDF first. You do not need to install anything, and a result in a minute or less shows what the tool returns.

The repository's app.py launches the code in the mineru_space folder. At startup the launcher builds an isolated environment from MinerU's development head, checks it and switches to it only if the checks pass. A failed update keeps an existing compatible environment, and the previous runtime manifests are kept for rollback. Table inference stays on the CPU, while the torch models use CUDA through ZeroGPU.

You upload a document
The request waits in the queue
A CPU sidecar or ZeroGPU parses it
Results are prepared as Middle JSON and assets
The page shows the finished result

Good to know: The online copy allows at most 20 pages per request, and a cancelled request keeps its slot until native work finishes.

• License: Apache 2.0
• Cost to run: none named, and the online copy runs on ZeroGPU hardware
• Data it reaches: the documents you upload, plus the opendatalab/MinerU-4_models_torch models and a cached runtime on disk
• Not for: documents longer than 20 pages in one request

On Hugging Face, 13 November 2026

• Likes: 686
• Built with: gradio
• License: apache-2.0

Links

• Space: https://huggingface.co/spaces/opendatalab/MinerU
• Files: https://huggingface.co/spaces/opendatalab/MinerU/tree/main

Try it

Open the Space in your browser, or copy it to your own Hugging Face account from its page. Nothing in this kit runs by itself.