README.md

---
tags:
- ocr
- document-processing
- paddleocr-vl
- ocr
- uv-script
- generated
---

# Document Processing using PaddleOCR-VL (OCR mode)

This dataset contains OCR results from images in [minhpvo/ocr-input](https://huggingface.co/datasets/minhpvo/ocr-input) using PaddleOCR-VL, an ultra-compact 0.9B OCR model.

## Processing Details

- **Source Dataset**: [minhpvo/ocr-input](https://huggingface.co/datasets/minhpvo/ocr-input)
- **Model**: [PaddlePaddle/PaddleOCR-VL](https://huggingface.co/PaddlePaddle/PaddleOCR-VL)
- **Task Mode**: `ocr` - General text extraction to markdown format
- **Number of Samples**: 13
- **Processing Time**: 1.9 min
- **Processing Date**: 2026-02-06 17:58 UTC

### Configuration

- **Image Column**: `image`
- **Output Column**: `paddleocr_ocr`
- **Dataset Split**: `train`
- **Batch Size**: 16
- **Smart Resize**: Enabled
- **Max Model Length**: 8,192 tokens
- **Max Output Tokens**: 4,096
- **Temperature**: 0.0
- **GPU Memory Utilization**: 80.0%

## Model Information

PaddleOCR-VL is a state-of-the-art, resource-efficient model tailored for document parsing:
- 🎯 **Ultra-compact** - Only 0.9B parameters (smallest OCR model)
- 📝 **OCR mode** - General text extraction
- 📊 **Table mode** - HTML table recognition
- 📐 **Formula mode** - LaTeX mathematical notation
- 📈 **Chart mode** - Structured chart analysis
- 🌍 **Multilingual** - Support for multiple languages
- ⚡ **Fast** - Quick initialization and inference
- 🔧 **ERNIE-4.5 based** - Different architecture from Qwen models

### Task Modes

- **OCR**: Extract text content to markdown format
- **Table Recognition**: Extract tables to HTML format
- **Formula Recognition**: Extract mathematical formulas to LaTeX
- **Chart Recognition**: Analyze and describe charts/diagrams

## Dataset Structure

The dataset contains all original columns plus:
- `paddleocr_ocr`: The extracted content based on task mode
- `inference_info`: JSON list tracking all OCR models applied to this dataset

## Usage

```python
from datasets import load_dataset
import json

# Load the dataset
dataset = load_dataset("{output_dataset_id}", split="train")

# Access the extracted content
for example in dataset:
    print(example["paddleocr_ocr"])
    break

# View all OCR models applied to this dataset
inference_info = json.loads(dataset[0]["inference_info"])
for info in inference_info:
    print(f"Task: {info['task_mode']} - Model: {info['model_id']}")
```

## Reproduction

This dataset was generated using the [uv-scripts/ocr](https://huggingface.co/datasets/uv-scripts/ocr) PaddleOCR-VL script:

```bash
uv run https://huggingface.co/datasets/uv-scripts/ocr/raw/main/paddleocr-vl.py \
    minhpvo/ocr-input \
    <output-dataset> \
    --task-mode ocr \
    --image-column image \
    --batch-size 16 \
    --max-model-len 8192 \
    --max-tokens 4096 \
    --gpu-memory-utilization 0.8
```

## Performance

- **Model Size**: 0.9B parameters (smallest among OCR models)
- **Processing Speed**: ~0.11 images/second
- **Architecture**: NaViT visual encoder + ERNIE-4.5-0.3B language model

Generated with 🤖 [UV Scripts](https://huggingface.co/uv-scripts)