openbmb/ minicpm-ocr1.0

109 Downloads Updated 11 hours ago

An open-source OCR solution for fast, accurate document parsing with correctness assessment.

vision

ollama run openbmb/minicpm-ocr1.0

curl http://localhost:11434/api/chat \
  -d '{
    "model": "openbmb/minicpm-ocr1.0",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

from ollama import chat

response = chat(
    model='openbmb/minicpm-ocr1.0',
    messages=[{'role': 'user', 'content': 'Hello!'}],
)
print(response.message.content)

import ollama from 'ollama'

const response = await ollama.chat({
  model: 'openbmb/minicpm-ocr1.0',
  messages: [{role: 'user', content: 'Hello!'}],
})
console.log(response.message.content)

Models

Name

12 models

Size / Usage

Context

Input

minicpm-ocr1.0:latest

1.4GB · 256K context window · Text, Image · 11 hours ago

minicpm-ocr1.0:latest

1.4GB

256K

Text, Image

minicpm-ocr1.0:q4_0

1.4GB · 256K context window · Text, Image · 11 hours ago

minicpm-ocr1.0:q4_0

1.4GB

256K

Text, Image

minicpm-ocr1.0:q4_1

1.4GB · 256K context window · Text, Image · 11 hours ago

minicpm-ocr1.0:q4_1

1.4GB

256K

Text, Image

minicpm-ocr1.0:q4_K_S

1.4GB · 256K context window · Text, Image · 11 hours ago

minicpm-ocr1.0:q4_K_S

1.4GB

256K

Text, Image

minicpm-ocr1.0:q4_K_M

1.4GB · 256K context window · Text, Image · 11 hours ago

minicpm-ocr1.0:q4_K_M

1.4GB

256K

Text, Image

minicpm-ocr1.0:q5_0

1.5GB · 256K context window · Text, Image · 11 hours ago

minicpm-ocr1.0:q5_0

1.5GB

256K

Text, Image

minicpm-ocr1.0:q5_1

1.5GB · 256K context window · Text, Image · 12 hours ago

minicpm-ocr1.0:q5_1

1.5GB

256K

Text, Image

minicpm-ocr1.0:q5_K_S

1.5GB · 256K context window · Text, Image · 12 hours ago

minicpm-ocr1.0:q5_K_S

1.5GB

256K

Text, Image

minicpm-ocr1.0:q5_K_M

1.5GB · 256K context window · Text, Image · 12 hours ago

minicpm-ocr1.0:q5_K_M

1.5GB

256K

Text, Image

minicpm-ocr1.0:q6_K

1.5GB · 256K context window · Text, Image · 12 hours ago

minicpm-ocr1.0:q6_K

1.5GB

256K

Text, Image

minicpm-ocr1.0:q8_0

1.7GB · 256K context window · Text, Image · 12 hours ago

minicpm-ocr1.0:q8_0

1.7GB

256K

Text, Image

minicpm-ocr1.0:f16

2.4GB · 256K context window · Text, Image · 12 hours ago

minicpm-ocr1.0:f16

2.4GB

256K

Text, Image

Readme

MiniCPM-OCR 1.0

GitHub · Hugging Face · ModelScope

MiniCPM-OCR 1.0 (1.3B) is the first OCR-specific model in the MiniCPM-V/o family. It parses Chinese and English documents and PDFs - text, tables, formulas, charts, seals, and key information - and holds up on scans, photographs, curved pages, and recaptured documents.

Notable features include:

🏆 SOTA document parsing. 93.68 on Real5-OmniDocBench, with leading results on OmniDocBench v1.6, Wild-OmniDocBench, and olmOCR-Bench.
🔍 General recognition and KIE. 54.7 on OCRBench v2, plus key information extraction straight from an image.
⚡ Efficient. 1.2 pages/s on a single RTX 5090, about 1.49x PaddleOCR-VL-1.6 and 3.62x MinerU2.5 Pro.
🛠️ Day-one Ollama support, alongside vLLM, SGLang, and llama.cpp.

Available tags

Tag	Quantization	Size	Notes
`latest` / `q4_k_m`	Q4_K_M	1.4 GB	recommended default
`q8_0`	Q8_0	1.7 GB	higher quality
`f16`	F16	2.4 GB	full precision

Sizes include the vision projector.

Quick start

”`bash ollama run openbmb/minicpm-ocr1.0 “Text Recognition: ./page.png”

# MiniCPM-OCR 1.0

<p align="center">
  <a href="https://github.com/OpenBMB/MiniCPM-OCR">GitHub</a> &nbsp;·&nbsp;
  <a href="https://huggingface.co/openbmb/MiniCPM-OCR-1.0">Hugging Face</a> &nbsp;·&nbsp;
  <a href="https://modelscope.cn/models/OpenBMB/MiniCPM-OCR-1.0">ModelScope</a>
</p>

**MiniCPM-OCR 1.0** (1.3B) is the first OCR-specific model in the MiniCPM-V/o family. It parses
Chinese and English documents and PDFs - text, tables, formulas, charts, seals, and key
information - and holds up on scans, photographs, curved pages, and recaptured documents.

Notable features include:

- 🏆 **SOTA document parsing.** 93.68 on Real5-OmniDocBench, with leading results on OmniDocBench
  v1.6, Wild-OmniDocBench, and olmOCR-Bench.
- 🔍 **General recognition and KIE.** 54.7 on OCRBench v2, plus key information extraction
  straight from an image.
- ⚡ **Efficient.** 1.2 pages/s on a single RTX 5090, about 1.49x PaddleOCR-VL-1.6 and 3.62x
  MinerU2.5 Pro.
- 🛠️ **Day-one Ollama support,** alongside vLLM, SGLang, and llama.cpp.

## Available tags

| Tag | Quantization | Size | Notes |
|:---|:---|:---|:---|
| `latest` / `q4_k_m` | Q4_K_M | 1.4 GB | recommended default |
| `q8_0` | Q8_0 | 1.7 GB | higher quality |
| `f16` | F16 | 2.4 GB | full precision |

Sizes include the vision projector.

## Quick start

```bash
ollama run openbmb/minicpm-ocr1.0 "Text Recognition: ./page.png"

Paste, drop or click to upload images (.png, .jpeg, .jpg, .svg, .gif)