Urdu_spelling_correction

Urdu Spelling Correction using DistilBERT

This repository contains a fine-tuned multilingual DistilBERT model for Urdu real-word spelling correction.

Model

• Architecture: DistilBERT (Multilingual)
• Task: Token Classification
• Language: Urdu
• Framework: Hugging Face Transformers

Training

The model was fine-tuned on aligned noisy/corrected Urdu text to predict the correct token using sentence context.

Usage

from transformers import AutoTokenizer
from transformers import AutoModelForTokenClassification

tokenizer = AutoTokenizer.from_pretrained(
    "HuzaifaQ-12/Urdu_spelling_correction"
)

model = AutoModelForTokenClassification.from_pretrained(
    "HuzaifaQ-12/Urdu_spelling_correction"
)

Project Repository

The complete project, methodology, preprocessing pipeline, report, and source code are available on GitHub:

https://github.com/HuzaifaQ-1/Urdu-Spelling-Correction

https://github.com/YourUsername/YourRepository

Notes

This repository contains only the trained model weights and tokenizer files.

For implementation details, dataset preparation, evaluation, and methodology, see the GitHub repository.