Modelpublic

Chhagan005/Chhagan_ML-VL-OCR-v1

sourceHugging Faceupdated 7mo agoView on Hugging Face

0likes16downloads

Model Card

Chhagan_ML-VL-OCR-v1

Multilingual OCR model for government ID cards, fine-tuned with LoRA on Qwen2.5-VL-3B.

Supported Documents

—🇴🇲 Oman Resident Card
—🇦🇪 UAE Identity Card
—🇸🇦 Saudi National ID
—🇮🇳 Aadhaar Card, PAN Card, Passport
—🇿🇦 South Africa ID

Languages

English, Arabic (عربي), Hindi (हिन्दी), Urdu (اردو)

Usage

Option 1: Load as LoRA Adapter (~74MB download)

python

from transformers import Qwen2_5_VLForConditionalGeneration, AutoProcessor
from peft import PeftModel

base = Qwen2_5_VLForConditionalGeneration.from_pretrained("Qwen/Qwen2.5-VL-3B-Instruct")
model = PeftModel.from_pretrained(base, "Chhagan005/Chhagan_ML-VL-OCR-v1")
processor = AutoProcessor.from_pretrained("Chhagan005/Chhagan_ML-VL-OCR-v1")

Option 2: Load as Standalone Model (~4.4GB download)

python

from transformers import Qwen2_5_VLForConditionalGeneration, AutoProcessor

model = Qwen2_5_VLForConditionalGeneration.from_pretrained(
    "Chhagan005/Chhagan_ML-VL-OCR-v1",
    torch_dtype=torch.float16, device_map="auto"
)
processor = AutoProcessor.from_pretrained("Chhagan005/Chhagan_ML-VL-OCR-v1")

Training

—Base: Qwen/Qwen2.5-VL-3B-Instruct
—Method: LoRA (r=32, α=64)
—Data: Synthetic IDs + ANETAC Arabic names + Ara-Eng Parallel Corpus
—Focus: OCR extraction from government identity documents