davanstrien/falcon-ocr-test-nocompile

---
tags:
- ocr
- document-processing
- falcon-ocr
- plain
- uv-script
- generated
---

# Document Processing using Falcon OCR (plain mode)

This dataset contains OCR results from images in [davanstrien/ufo-ColPali](https://huggingface.co/datasets/davanstrien/ufo-ColPali) using [Falcon OCR](https://huggingface.co/tiiuae/Falcon-OCR), a 0.3B early-fusion vision-language model.

## Processing Details

- **Source Dataset**: [davanstrien/ufo-ColPali](https://huggingface.co/datasets/davanstrien/ufo-ColPali)
- **Model**: [tiiuae/Falcon-OCR](https://huggingface.co/tiiuae/Falcon-OCR)
- **Task Mode**: `plain` - Full-page text extraction
- **Number of Samples**: 20
- **Processing Time**: 9.3 min
- **Processing Date**: 2026-04-07 20:27 UTC

### Configuration

- **Image Column**: `image`
- **Dataset Split**: `train`
- **Max Output Tokens**: 2,048
- **Backend**: Transformers

## Model Information

Falcon OCR is a compact early-fusion model that processes image patches and text tokens
in a shared Transformer. Key results:
- 80.3% on olmOCR benchmark
- 88.64% on OmniDocBench
- 87.1% on multi-column documents (best in class)
- 90.3% on tables (best in class)

## Reproduction

```bash
uv run https://huggingface.co/datasets/uv-scripts/ocr/raw/main/falcon-ocr.py \
davanstrien/ufo-ColPali \
<output-dataset> \
--task-mode plain \
--image-column image
```

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