README.md

---
title: Ocr Entropy
emoji: 🏢
colorFrom: blue
colorTo: purple
sdk: gradio
sdk_version: 6.4.0
python_version: "3.10"
app_file: app.py
pinned: false
license: mit
short_description: Calculating the probabilities and entropy of OCR output
---

# OCR Confidence Visualization

Extract text from document images with token-level confidence visualization.

## Features

- **Token Streaming**: Watch text appear token-by-token as the model generates
- **Confidence Colors**: Each token is colored based on model confidence:
  - Blue (>99%): Very high confidence
  - Dark Green (>95%): High confidence
  - Light Green (>85%): Good confidence
  - Amber (>70%): Moderate confidence
  - Red (>50%): Low confidence
  - Purple (<=50%): Very low confidence
- **Token Alternatives**: Click any token to see top alternative predictions with probabilities

## Model

Uses [Nanonets-OCR2-3B](https://huggingface.co/nanonets/Nanonets-OCR2-3B), a Qwen2.5-VL-3B fine-tune optimized for document OCR.

## Usage

1. Upload a document image (JPG, PNG, etc.)
2. Click "Transcribe"
3. Watch tokens stream with confidence coloring
4. Click any token to see alternative predictions

## Technical Details

- Extracts logprobs from each generated token
- Converts logprobs to probabilities via softmax
- Top-k alternatives stored for each token (k=20)
- ZeroGPU compatible for HuggingFace Spaces deployment