README.md

---
language:
- en
tags:
- vision-language
- vqa
- text-to-image-evaluation
license: mit
---

# Tiny Random VQAScore Model

This is a tiny random version of the VQAScore architecture for educational and testing purposes.

## Model Architecture

- **Vision Encoder**: Tiny CNN + Transformer (64 hidden size)
- **Language Model**: Tiny Transformer (256 hidden size)
- **Multimodal Projector**: MLP with 256 → 128 → 64 → 1

## Usage

```python
from create_tiny_vqa_model import TinyVQAScore

# Load the model
model = TinyVQAScore(device="cpu")

# Score an image
from PIL import Image
image = Image.open("your_image.jpg")
score = model.score(image, "What is shown in this image?")
print(f"VQA Score: {score}")
```

## Model Size

- **Parameters**: ~50K (vs ~11B for the original XXL model)
- **Memory**: ~200KB (vs ~22GB for the original XXL model)

## Disclaimer

This is a randomly initialized model for testing and educational purposes. It is not trained and will not produce meaningful VQA results.