---
license: apache-2.0
tags:
- vision
- ocr
- document-analysis
- multimodal
- qwen2vl
- gguf
- lm-studio
model_type: vision-language
architecture: qwen2vl
quantization: q8_0
languages:
- en
- zh
pipeline_tag: image-text-to-text
---

# MonkeyOCR-pro-1.2B Vision GGUF

A high-performance **vision-language model** specialized for **Optical Character Recognition (OCR)** and **document analysis**. This repository contains GGUF format models optimized for both general use and **LM Studio** compatibility.

## 🎯 Model Capabilities

- ✅ **Vision-Language Processing**: Understand and process images with text
- ✅ **OCR (Optical Character Recognition)**: Extract text from images and documents  
- ✅ **Document Structure Analysis**: Analyze layout, tables, and formatting
- ✅ **Multi-turn Conversations**: Chat about images and documents
- ✅ **Multiple Languages**: Support for English, Chinese, and more

## 📁 Files Included

### Standard GGUF Models
- **`MonkeyOCR-pro-1.2B-Recognition.gguf`** (1.26 GB) - Complete vision-enabled model
- **`MonkeyOCR-pro-1.2B.gguf`** (1.26 GB) - Alternative version

### LM Studio Optimized
- **`MonkeyOCR-pro-1.2B-Text-LMStudio.gguf`** (1.23 GB) - Text model for LM Studio
- **`mmproj-MonkeyOCR-pro-1.2B-Vision-LMStudio.gguf`** (1.25 GB) - Vision projection for LM Studio

## 🎨 LM Studio Setup (Recommended)

### Step 1: Download Files
Download both LM Studio files:
- `MonkeyOCR-pro-1.2B-Text-LMStudio.gguf`
- `mmproj-MonkeyOCR-pro-1.2B-Vision-LMStudio.gguf`

### Step 2: Install in LM Studio
1. Copy both files to your LM Studio models directory
2. Keep them in the same folder
3. Open LM Studio and load the text model: `MonkeyOCR-pro-1.2B-Text-LMStudio.gguf`
4. LM Studio will automatically detect and load the vision projection

### Step 3: Enable Vision
1. Start a new chat session
2. Look for the camera/image icon in the chat interface  
3. Upload an image and start asking questions!

## 🔧 Usage Examples

### OCR Tasks
```
User: [uploads image] Extract all text from this document
Assistant: I can see text in the image. Here's what I extracted:
[Provides accurate OCR results]
```

### Document Analysis
```
User: [uploads form] Analyze the structure of this form
Assistant: This appears to be a [form type] with the following structure:
- Header section with title
- Main content area with fields for...
- Footer with signature area
```

### Vision Q&A
```
User: [uploads chart] What does this chart show?
Assistant: This chart displays [detailed analysis of the visual content]
```

## ⚙️ Technical Specifications

- **Architecture**: Qwen2.5-VL (qwen2vl)
- **Model Size**: 1.2B parameters
- **Quantization**: Q8_0 (8-bit)
- **Context Length**: 8,196 tokens
- **Vision Encoder**: CLIP-based projection
- **Input Support**: Images, text, multimodal conversations

## 🔧 Configuration Details

The model includes proper vision token configuration:
- Image Token ID: 151655 (`<|image_pad|>`)
- Video Token ID: 151656 (`<|video_pad|>`)
- Vision Start: 151652 (`<|vision_start|>`)
- Vision End: 151653 (`<|vision_end|>`)

## 🚀 Performance

- **Fast inference** with Q8_0 quantization
- **High accuracy** OCR capabilities
- **Efficient memory usage** (~1.2-1.3 GB VRAM)
- **Multi-language support** for global documents

## 📝 License

Apache 2.0 - Free for commercial and research use

## 🙏 Credits

Based on the MonkeyOCR model architecture with optimizations for GGUF format and LM Studio compatibility.

---

**For best vision performance, use the LM Studio optimized files!** 🎯