pxoooArt/qwen2.5-vl-7b-arabic-invoice-extractor-v2

Qwen2.5-VL 7B Arabic Invoice Extractor

Model Description

This is a fine-tuned version of Qwen2.5-VL-7B-Instruct specifically trained for extracting structured data from Arabic invoices. The model uses LoRA (Low-Rank Adaptation) to efficiently fine-tune the base model for this specialized task.

Training Details

• Base Model: Qwen/Qwen2.5-VL-7B-Instruct
• Training Method: LoRA fine-tuning with Unsloth
• Dataset: Arabic invoice images
• Target Fields:
  • date: Invoice date in YYYY-MM-DD format
  • client: Client/buyer name (preferring Arabic names)
  • referenceNumber: Invoice number
  • totalAmount: Amount before VAT (float)

Usage

from transformers import AutoModelForVision2Seq, AutoProcessor
from peft import PeftModel
import torch
from PIL import Image

# Load base model
base_model = AutoModelForVision2Seq.from_pretrained(
    "Qwen/Qwen2.5-VL-7B-Instruct",
    torch_dtype=torch.float16,
    device_map="auto",
    trust_remote_code=True
)

# Load LoRA adapter
model = PeftModel.from_pretrained(base_model, "pxoooArt/qwen2.5-vl-7b-arabic-invoice-extractor-v2")
processor = AutoProcessor.from_pretrained("Qwen/Qwen2.5-VL-7B-Instruct", trust_remote_code=True)

# Process image
image = Image.open("invoice.jpg")
# ... (see handler.py for complete inference code)

Expected Output Format

{
  "date": "2024-03-15",
  "client": "عبايات النور", 
  "referenceNumber": "INV-2024-001",
  "totalAmount": 1250.50
}

Performance

• Optimized for Arabic business documents
• Handles various invoice formats common in Saudi Arabia
• Accurate date parsing with middle-number-as-month logic
• Reliable extraction of Arabic client names

Handler Code

The model includes a complete handler for deployment. See the repository files for handler.py.

License

This model inherits the license from the base Qwen2.5-VL model.