Repolode
Submit
OpenBMB

OpenBMB/MiniCPM-OCR

MiniCPM-OCR is a family of open-source document parsing models, plus a companion model that scores whether OCR results are correct.

Open-source alternative to:ABBYY FineReader

View on GitHub
Stars
846
Forks
62
Open issues
1
Last commit
yesterday

What is it?

MiniCPM-OCR is a set of models for parsing Chinese and English documents and PDFs into text, tables, and formulas. MiniCPM-OCR 1.0 (1.3B parameters) handles challenging inputs such as scans, photographs, curved and skewed pages, and recaptured documents, and it also supports key information extraction.

MiniCPM-OCR Jev is a separate 2B decision model. Given an image and an OCR result, it returns a quality score from 0 to 1 at the region level, which helps teams pick reliable outputs and locate content that needs human review. It is designed to work with the output of different parsing models.

The project supports inference through vLLM, SGLang, llama.cpp, and Ollama, with BF16, GGUF, AWQ, and GPTQ weights published on Hugging Face and ModelScope. It also includes a pip SDK, a hosted API, and fine-tuning tutorials for LLaMA-Factory and Swift.

Key features

  • 01Parses Chinese and English documents and PDFs into Markdown, tables, and formulas.
  • 02Handles scans, photographs, curved pages, and recaptured documents.
  • 03Supports key information extraction alongside text recognition.
  • 04Scores OCR results per region to flag content that needs review.
  • 05Runs on vLLM, SGLang, llama.cpp, and Ollama, with quantized weights available.
  • 06Offers a pip SDK, a hosted API, and fine-tuning guides.

Use cases

  • Converting scanned contracts, financial statements, or reports into searchable Markdown.
  • Extracting structured fields from receipts or forms for downstream processing.
  • Filtering OCR output so only low-scoring regions go to human review.
  • Fine-tuning a document parser for a specific domain or document layout.

Quick start

  1. 1

    Install the MiniCPM-OCR pip SDK

  2. 2

    Follow the MiniCPM-OCR 1.0 local deployment tutorial to run the models

  3. 3

    Or use the Free API for recognition without local hardware

Things to know

  • Running the models locally requires your own hardware, and the README does not list minimum specs.
  • Some top benchmark claims are reported by the project itself and have not been independently verified here.
  • Hosted Free API use depends on a third-party platform account and its terms.

In collections

Run LLMs locallyPDF, OCR & document tools

Topics

#ocr#document-parsing#vision-language-model#pdf#key-information-extraction#llm-inference#fine-tuning#image2text#minicpm-v

FAQ

What is MiniCPM-OCR?
MiniCPM-OCR is a family of open-source document parsing models, plus a companion model that scores whether OCR results are correct. MiniCPM-OCR is a set of models for parsing Chinese and English documents and PDFs into text, tables, and formulas. MiniCPM-OCR 1.0 (1.3B parameters) handles challenging inputs such as scans, photographs, curved and skewed pages, and recaptured documents, and it also supports key information extraction.
Is MiniCPM-OCR free to use? What license does it have?
MiniCPM-OCR is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
Is MiniCPM-OCR still maintained?
Actively maintained. The last commit to MiniCPM-OCR was on Nov 6, 2026.
What can I use MiniCPM-OCR for?
Converting scanned contracts, financial statements, or reports into searchable Markdown. Extracting structured fields from receipts or forms for downstream processing. Filtering OCR output so only low-scoring regions go to human review. Fine-tuning a document parser for a specific domain or document layout.
How do I install MiniCPM-OCR?
Install the MiniCPM-OCR pip SDK Follow the MiniCPM-OCR 1.0 local deployment tutorial to run the models Or use the Free API for recognition without local hardware
Where can I get MiniCPM-OCR?
The source code and installation instructions are on GitHub: https://github.com/OpenBMB/MiniCPM-OCR

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects