---
library_name: transformers
tags:
- mlx
base_model:
- Qwen/Qwen2.5-VL-3B-Instruct
---


# 📸 Adasah - Qwen 2.5 3B (4-bit) Fine-tuned on Arabic Photo Q&A and Descriptions

## Demo Video:

<video controls autoplay src="https://cdn-uploads.huggingface.co/production/uploads/630535e0c7fed54edfaa1a75/Jv2jNMCu98Fe9BnZ3wKJ9.mp4" width="400px"></video>

## Adasah - IOS App 
[App Store](https://apps.apple.com/us/app/adasah/id6745417467)
### Warning - The app downloads a 2GB model, so it takes some time for the first time.

## Forked from Huggingsnap
https://github.com/huggingface/HuggingSnap

**Model Name**: `Adasah`  
**Base Model**: [Qwen/Qwen2.5-VL-3B-Instruct](https://huggingface.co/Qwen/Qwen2.5-VL-3B-Instruct)
**Quantization**: 4-bit (GGUF)  
**Platform**: iOS (mobile-compatible)  
**Language**: Arabic (translated from English)  
**Use Case**: Arabic Visual Q&A and Photo Description Understanding

---

## 🧠 Model Overview

**Adasah** is a fine-tuned variant of the Qwen 2.5 3B base model, optimized for Arabic language understanding in visual contexts. The model was trained on a custom dataset consisting of **English visual question-answer pairs and photo descriptions translated into Arabic**, allowing it to:

- Answer Arabic questions about images
- Generate Arabic descriptions of visual content
- Serve as a mobile assistant for Arabic-speaking users

The model is quantized to **4-bit** to ensure smooth on-device performance on **iOS apps**.


## 📱 Mobile Optimization

The model is quantized using **4-bit precision** to make it lightweight and suitable for **on-device inference** in:

- **iOS apps**
- **Offline-first mobile experiences**
- **Arabic language educational or accessibility tools**

---

## Use with mlx

```bash
pip install -U mlx-vlm
```

```bash
python -m mlx_vlm.generate --model NAMAA-Space/Adasah-QA-0.1-3B-Instruct-merged-4bits --max-tokens 100 --temperature 0.0 --prompt "Describe this image." --image <path_to_image>
```