OpenDoc-0.1B: Ultra-Lightweight Document Parsing System with 0.1B Parameters
Introduction
OpenDoc-0.1B is an ultra-lightweight document parsing system featuring only 0.1 billion parameters. It operates through a sophisticated two-stage pipeline: first, it utilizes PP-DocLayoutV2 for precise layout analysis; second, it employs an enhanced, in-house UniRec-0.1B model for the unified recognition of text, formulas, and tables. While the original version of UniRec-0.1B focused solely on text and formulas, this rebuilt iteration integrates comprehensive table recognition capabilities. Supporting both Chinese and English document parsing, OpenDoc-0.1B achieves an impressive 90.57% score on OmniDocBench (v1.5), demonstrating superior performance that outrivals many large-scale multimodal document parsing models.
Quick Start
Requirements
conda create -n openocr python=3.10
conda activate openocr
git clone https://github.com/Topdu/OpenOCR.git
cd OpenOCR
pip install -r requirements.txt
python -m pip install paddlepaddle-gpu==3.2.0 -i https://www.paddlepaddle.org.cn/packages/stable/cu126/
python -m pip install paddlex
pip install torch==2.6.0 torchvision==0.21.0 torchaudio==2.6.0 --index-url https://download.pytorch.org/whl/cu126
pip install transformers==4.49.0
Download UniRec-0.1B model
# download model from modelscope
modelscope download topdktu/unirec-0.1b --local_dir ./unirec-0.1b
# or download model from huggingface
huggingface-cli download topdu/unirec-0.1b --local-dir ./unirec-0.1b
Inference
# cpu
python tools/infer_doc.py --input_path ../doc_img_or_pdf --output_path ./output --gpus -1
# gpu
python tools/infer_doc.py --input_path ../doc_img_or_pdf --output_path ./output --gpus 0
# multi gpu
python tools/infer_doc.py --input_path ../doc_img_or_pdf --output_path ./output --gpus 0,1,2,3,4,5,6,7
Local Demo
pip install gradio==4.20.0
python demo_opendoc.py
Citation
If you find our method useful for your research, please cite:
@article{du2025unirec,
title={UniRec-0.1B: Unified Text and Formula Recognition with 0.1B Parameters},
author={Yongkun Du and Zhineng Chen and Yazhen Xie and Weikang Bai and Hao Feng and Wei Shi and Yuchen Su and Can Huang and Yu-Gang Jiang},
journal={arXiv preprint arXiv:2512.21095},
year={2025}
}