xiaomi-research/Xiaomi-OCR-0

What is it?

Key features

• 01Parses document pages into Markdown, with tables as HTML and formulas in LaTeX.
• 02Extracts requested fields from documents as JSON.
• 03Answers OCR-centric questions about document images.
• 04Provides an MCP service and Agent Skill for local agent integration.
• 05Offers a local browser demo for page, region, KIE, VQA and PDF parsing.
• 06Runs on SGLang or vLLM, with experimental Ollama support.

Use cases

• Converting scanned or digital PDFs into Markdown for downstream processing.
• Extracting structured fields such as invoice or form data as JSON.
• Asking questions about text, charts or tables in document images.
• Running a local OCR service that an AI agent can call through MCP.

Quick start

1. 1

   Clone the repository

   git clone https://github.com/SeerRay-Lab/Xiaomi-OCR-0.git
2. 2

   Start the vLLM inference server (keep it running)

   vllm serve SeerRay-Lab/Xiaomi-OCR-0 --host 127.0.0.1 --port 8000 --max-model-len 16384
3. 3

   In the repository directory, create a virtual environment and install requirements

   python3 -m venv .venv source .venv/bin/activate python -m pip install -r requirements.txt
4. 4

   Start the browser demo, then open http://127.0.0.1:8787

   python demo/server.py --port 8787

Things to know

• !Needs a compatible GPU and a SGLang or vLLM runtime installed separately, which have hardware-specific requirements.
• !Region parsing needs extra PaddlePaddle and PaddleX dependencies.
• !The Ollama path is experimental and its accuracy is not validated against the benchmarks.

In collections

Topics

FAQ

What is Xiaomi-OCR-0?

A unified 0.8B vision-language model for document parsing, key information extraction and OCR-centric visual question answering. Xiaomi-OCR-0 is a 0.8B vision-language model built on Qwen3.5-0.8B and trained on an OCR-centric corpus. It turns document page images into structured Markdown, with tables converted to HTML and formulas written in LaTeX. It also extracts requested fields as JSON and answers questions about text in document images.

Is Xiaomi-OCR-0 free to use? What license does it have?

Xiaomi-OCR-0 is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.

Is Xiaomi-OCR-0 still maintained?

Actively maintained. The last commit to Xiaomi-OCR-0 was on Oct 29, 2026.

What can I use Xiaomi-OCR-0 for?

Converting scanned or digital PDFs into Markdown for downstream processing. Extracting structured fields such as invoice or form data as JSON. Asking questions about text, charts or tables in document images. Running a local OCR service that an AI agent can call through MCP.

How do I install Xiaomi-OCR-0?

Clone the repository: git clone https://github.com/SeerRay-Lab/Xiaomi-OCR-0.git Start the vLLM inference server (keep it running): vllm serve SeerRay-Lab/Xiaomi-OCR-0 --host 127.0.0.1 --port 8000 --max-model-len 16384 In the repository directory, create a virtual environment and install requirements: python3 -m venv .venv source .venv/bin/activate python -m pip install -r requirements.txt Start the browser demo, then open http://127.0.0.1:8787: python demo/server.py --port 8787

Where can I get Xiaomi-OCR-0?

The source code and installation instructions are on GitHub: https://github.com/xiaomi-research/Xiaomi-OCR-0