RAFAY-484/Urdu-Punjabi-V2

Urdu-Punjabi Language Learning Model V2

This is a fine-tuned XLM-RoBERTa model for:

• Answer Scoring: Evaluating user responses in Urdu and Pakistani Punjabi
• Grammar Checking: Validating sentence structure
• Translation Validation: Checking translation accuracy

Model Details

• Base Model: xlm-roberta-base
• Languages: Urdu (Nastaliq), Pakistani Punjabi (Shahmukhi), English
• Task: Regression (score 0.0 to 1.0)
• Fine-tuned on: Custom vocabulary dataset (1000 words)

Dataset

The model was trained on a comprehensive custom dataset with:

• 20 Chapters (10 Urdu + 10 Punjabi)
• 50 words per chapter = 1000 vocabulary items
• 100 Quiz MCQ Questions (5 per chapter)
• 100 User Input Questions (5 per chapter)
• 1000+ Grammar Examples (sentence-translation pairs)

Chapter Topics:

1. Greetings & Polite Expressions (سلام و آداب)
2. Family & Relationships (خاندان اور رشتے)
3. Food & Dining (کھانا اور خوراک)
4. Numbers & Counting (گنتی اور اعداد)
5. Places & Locations (جگہیں اور مقامات)
6. Shopping & Money (خریداری اور پیسے)
7. Emotions & Feelings (جذبات اور احساسات)
8. Weather & Nature (موسم اور فطرت)
9. Body Parts & Health (جسم کے اعضاء اور صحت)
10. Education & Learning (تعلیم اور سیکھنا)

Usage

Python

from transformers import XLMRobertaTokenizer, XLMRobertaForSequenceClassification
import torch

# Load model
model_name = "RAFAY-484/Urdu-Punjabi-V2"
tokenizer = XLMRobertaTokenizer.from_pretrained(model_name)
model = XLMRobertaForSequenceClassification.from_pretrained(model_name)

# Score an answer
expected = "خوش"  # Happy in Urdu/Punjabi
user_input = "خوش"

inputs = tokenizer(expected, user_input, return_tensors="pt")
with torch.no_grad():
    outputs = model(**inputs)
    score = torch.sigmoid(outputs.logits).item()

print(f"Score: {score:.3f}")  # Output: Score: 0.95+

API Usage (Flutter/Mobile App)

Future<double> scoreAnswer(String expected, String userInput) async {
  final response = await http.post(
    Uri.parse('https://api-inference.huggingface.co/models/RAFAY-484/Urdu-Punjabi-V2'),
    headers: {
      'Authorization': 'Bearer YOUR_HF_TOKEN',
      'Content-Type': 'application/json',
    },
    body: jsonEncode({
      'inputs': '$expected [SEP] $userInput',
    }),
  );
  
  final result = jsonDecode(response.body);
  return result[0]['score'];
}

Scoring Guide

Score Range	Meaning	Example
0.9 - 1.0	Exact match	Expected: خوش, User: خوش
0.7 - 0.9	Close match	Expected: السلام علیکم, User: السلام
0.4 - 0.7	Partial match	Expected: شکریہ, User: شکر
0.2 - 0.4	Related word	Expected: کتاب, User: book
0.0 - 0.2	Incorrect	Expected: پیار, User: نفرت

Vocabulary Samples

Urdu

Word	Translation	Pronunciation
شکریہ	Thank you	shukriya
خوش آمدید	Welcome	khush aamdeed
براہ کرم	Please	barah-e-karam

Pakistani Punjabi (Shahmukhi)

Word	Translation	Pronunciation
جی آیاں نوں	Welcome	ji aayan nu
ودھیا	Very good	wadiya
سوہنا	Beautiful	sohna

Important Note on Punjabi

⚠️ This model uses Pakistani Muslim Punjabi written in Shahmukhi script (Arabic-based). It does NOT include Sikh/Hindi Punjabi (Gurmukhi script) words.

All Punjabi vocabulary is authentic Pakistani Punjabi as spoken in Punjab, Pakistan.

Training Configuration

• Epochs: 2
• Batch Size: 32
• Learning Rate: 2e-05
• Max Length: 128
• Training Samples: 3386
• Validation Samples: 598

Use Cases

1. Language Learning Apps: Score user responses in vocabulary exercises
2. Quiz Systems: Validate answers in MCQ and fill-in-the-blank questions
3. Grammar Checking: Evaluate sentence correctness
4. Translation Apps: Verify translation accuracy

Citation

@misc{urdu-punjabi-v2-2024,
  author = {RAFAY-484},
  title = {Urdu-Punjabi Language Learning Model V2},
  year = {2024},
  publisher = {HuggingFace},
  howpublished = {\url{https://huggingface.co/RAFAY-484/Urdu-Punjabi-V2}}
}

License

MIT License - Free for educational and commercial use.