Qari-OCR-0.1-VL-2B-Instruct Model

Model Overview

This model is a fine-tuned version of unsloth/Qwen2-VL-2B-Instruct on an Arabic OCR dataset. It is optimized to perform Arabic Optical Character Recognition (OCR) for full-page text.

• It is described in detail in the paper QARI-OCR: High-Fidelity Arabic Text Recognition through Multimodal Large Language Model Adaptation.

Model Details

• Base Model: Qwen2 VL
• Fine-tuning Dataset: Arabic OCR dataset
• Objective: Extract full-page Arabic text with high accuracy
• Languages: Arabic
• Tasks: OCR (Optical Character Recognition)
• Dataset size: 5000 records
• Epochs: 1

Performance Evaluation

The model has been evaluated on standard OCR metrics, including Word Error Rate (WER), Character Error Rate (CER), and BLEU score.

Metrics

Model	WER ↓	CER ↓	BLEU ↑
Qari v0.1 Model	0.068	0.019	0.860
Qwen2 VL 2B	1.344	1.191	0.201
EasyOCR	0.908	0.617	0.152
Tesseract OCR	0.428	0.226	0.410

Key Results

• WER: 0.068 (93.2% word accuracy)
• CER: 0.019 (98.1% character accuracy)
• BLEU: 0.860

Performance Comparison

The Fine-Tuned Model outperforms other solutions with:

• 95% reduction in WER compared to Base Model
• 98% reduction in CER compared to Base Model
• 328% improvement in BLEU score compared to Base Model
• 84% lower WER than Tesseract OCR
• 92% lower WER than EasyOCR

Performance Comparison Charts

WER & CER Comparison

BLEU Score Comparison

Limitations

While the Arabic OCR model demonstrates strong performance under specific conditions, it has several limitations:

1. Font Dependency: The model was trained using a limited set of fonts (Almarai-Regular, Amiri-Regular, Cairo-Regular, Tajawal-Regular, and NotoNaskhArabic-Regular). As a result, its accuracy may degrade when processing text in other fonts, particularly decorative or stylized typefaces.

2. Font Size Restriction: Training was conducted with a fixed font size of 16. Variations in font size, especially very small or large text, may reduce recognition accuracy.

3. Diacritics Exclusion: The model does not support Arabic diacritics (Tashkeel). Text that relies on diacritics for disambiguation may not be correctly recognized.

4. Lack of Handwriting Support: The model is not trained to recognize handwritten text, limiting its applicability to printed documents only.

5. Full-Page Processing: The model was trained on full-page text recognition, which may impact its performance on segmented text, cropped sections, or text within complex layouts such as tables and multi-column formats.

These limitations should be considered when deploying the model in real-world applications to ensure optimal performance.

How to Use

Try Qari - Google Colab

You can load this model using the transformers and qwen_vl_utils library:

!pip install transformers qwen_vl_utils accelerate>=0.26.0 PEFT -U
!pip install -U bitsandbytes

from PIL import Image
from transformers import Qwen2VLForConditionalGeneration, AutoProcessor
import torch
import os
from qwen_vl_utils import process_vision_info



model_name = "NAMAA-Space/Qari-OCR-0.1-VL-2B-Instruct"
model = Qwen2VLForConditionalGeneration.from_pretrained(
                model_name,
                torch_dtype="auto",
                device_map="auto"
            )
processor = AutoProcessor.from_pretrained(model_name)
max_tokens = 2000

prompt = "Below is the image of one page of a document, as well as some raw textual content that was previously extracted for it. Just return the plain text representation of this document as if you were reading it naturally. Do not hallucinate."
image.save("image.png")

messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "image": f"file://{src}"},
            {"type": "text", "text": prompt},
        ],
    }
]
text = processor.apply_chat_template(
    messages, tokenize=False, add_generation_prompt=True
)
image_inputs, video_inputs = process_vision_info(messages)
inputs = processor(
    text=[text],
    images=image_inputs,
    videos=video_inputs,
    padding=True,
    return_tensors="pt",
)
inputs = inputs.to("cuda")
generated_ids = model.generate(**inputs, max_new_tokens=max_tokens)
generated_ids_trimmed = [
    out_ids[len(in_ids) :] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
]
output_text = processor.batch_decode(
    generated_ids_trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False
)[0]
os.remove(src)
print(output_text)

License

This model follows the licensing terms of the original Qwen2 VL model. Please review the terms before using it commercially.

Citation

If you use this model in your research, please cite:

@article{wasfy2025qari,
  title={QARI-OCR: High-Fidelity Arabic Text Recognition through Multimodal Large Language Model Adaptation},
  author={Wasfy, Ahmed and Nacar, Omer and Elkhateb, Abdelakreem and Reda, Mahmoud and Elshehy, Omar and Ammar, Adel and Boulila, Wadii},
  journal={arXiv preprint arXiv:2506.02295},
  year={2025}
}