MonkeyOCRv2-Parsing on CPU

This guide covers running MonkeyOCRv2 Document Parsing on CPU. For GPU inference, see Document Parsing in this repository.

CPU inference loads only the parsing checkpoint passed to -m / --model-path (MonkeyOCRv2-B-Parsing or MonkeyOCRv2-S-Parsing). Do not download MonkeyOCRv2-B-Parsing-DFlash for this guide: that draft model is used only for GPU vLLM speculative decoding.

---

Setup MonkeyOCRv2-Parsing(Windows)

1. Clone the repository

git clone https://github.com/Yuliang-Liu/MonkeyOCRv2.git
cd MonkeyOCRv2

2. Create virtual environment

py -3.11 -m venv .venv

3. Install dependencies

.\.venv\Scripts\python.exe -m pip install --upgrade pip
.\.venv\Scripts\python.exe -m pip install torch==2.5.1 torchvision==0.20.1 --index-url https://download.pytorch.org/whl/cpu
.\.venv\Scripts\python.exe -m pip install -r parsing\requirements.txt
.\.venv\Scripts\python.exe -m pip install transformers==4.57.1 accelerate==1.11.0 huggingface_hub qwen_vl_utils opencv-python einops

4. Download model weights

From ModelScope (faster in China):

.\.venv\Scripts\python.exe -m pip install modelscope
python download_model.py -t modelscope -n MonkeyOCRv2-B-Parsing # or MonkeyOCRv2-S-Parsing

Or from HuggingFace:

python download_model.py -n MonkeyOCRv2-B-Parsing # or MonkeyOCRv2-S-Parsing

---

Parse using CLI

cd .\parsing\
# Parse a single document
..\.venv\Scripts\python.exe cpu\parse_cpu.py `
    -i ..\images_test\vqa.png `
    -m ..\model_weight\MonkeyOCRv2-B-Parsing `
    -o output\test_cpu
# Parse a directory containing PDFs or images
..\.venv\Scripts\python.exe cpu\parse_cpu.py `
    -i ..\images_test `
    -m ..\model_weight\MonkeyOCRv2-B-Parsing `
    -o output\test_cpu

---

Serve with Web Demo

cd .\parsing\
..\.venv\Scripts\python.exe cpu\gradio_demo_cpu.py `
    --model-path ..\model_weight\MonkeyOCRv2-B-Parsing `
    --output-dir output\demo_cpu_outputs `
    --demo-server-name 127.0.0.1 `
    --demo-server-port 8891 `
    --demo-concurrency 1 `
    --page-max-inflight 1

---

Setup MonkeyOCRv2-Parsing(Linux)

1. Clone the repository

git clone https://github.com/Yuliang-Liu/MonkeyOCRv2.git
cd MonkeyOCRv2

2. Create virtual environment

conda create -n MonkeyOCRv2Parsing_CPU python=3.11 -y
conda activate MonkeyOCRv2Parsing_CPU

3. Install dependencies

pip install torch==2.5.1 torchvision==0.20.1 --index-url https://download.pytorch.org/whl/cpu
pip install -r parsing/requirements.txt
pip install transformers==4.57.1 accelerate==1.11.0 huggingface_hub qwen_vl_utils opencv-python einops

4. Download model weights

From ModelScope (faster in China):

pip install modelscope
python download_model.py -t modelscope -n MonkeyOCRv2-B-Parsing # or MonkeyOCRv2-S-Parsing

Or from HuggingFace:

python download_model.py -n MonkeyOCRv2-B-Parsing # or MonkeyOCRv2-S-Parsing

---

Parse using CLI

cd parsing
# Parse a single document
python cpu/parse_cpu.py \
    -i ../images_test/vqa.png \
    -m ../model_weight/MonkeyOCRv2-B-Parsing \
    -o output/test_cpu
# Parse a directory containing PDFs or images
python cpu/parse_cpu.py \
    -i ../images_test \
    -m ../model_weight/MonkeyOCRv2-B-Parsing \
    -o output/test_cpu

---

Serve with Web Demo

cd parsing
python cpu/gradio_demo_cpu.py \
    --model-path ../model_weight/MonkeyOCRv2-B-Parsing \
    --output-dir output/demo_cpu_outputs \
    --demo-server-name 127.0.0.1 \
    --demo-server-port 8891 \
    --demo-concurrency 1 \
    --page-max-inflight 1
# Show help messages
python cpu/gradio_demo_cpu.py -h