Model Card for Bart Urdu Summarizer
This model is designed to summarize Urdu text using the BART architecture, fine-tuned on a custom Urdu summarization dataset.
Model Details
Model Description
This model leverages the BART (Bidirectional and Auto-Regressive Transformers) architecture to perform Urdu text summarization. The model was fine-tuned on a headline-based Urdu dataset to generate concise and meaningful summaries. It is well-suited for tasks like news summarization, article summarization, and extracting key points from long texts.
Developed by: Mudasir692
Model type: BART
Language(s) (NLP): Urdu
License: MIT
Finetuned from model: facebook/bart-large
Model Sources
Repository:
https://huggingface.co/Mudasir692/bart-urdu-summarizer
Uses
Direct Use
This model is intended for generating concise summaries of Urdu text directly from input data.
Downstream Use
The model can be fine-tuned further for specific tasks involving Urdu summarization or adapted for multilingual summarization tasks.
Out-of-Scope Use
The model may not perform well on highly specialized domains or technical documents without additional fine-tuning. It is not suitable for generating summaries of text in languages other than Urdu.
Bias, Risks, and Limitations
The model may inherit biases from the training data, particularly in topics and vocabulary frequently represented in the dataset. The summaries may occasionally miss critical context or introduce ambiguities.
Recommendations
Users should validate the summaries in sensitive applications and consider fine-tuning or additional post-processing for domain-specific requirements.
How to Get Started with the Model
To get started with the model, use the following code snippet to load the model and tokenizer, input Urdu text, and generate concise summaries.
python
Copy code
import torch
from transformers import MBartForConditionalGeneration, MBart50Tokenizer
Load the tokenizer and model
tokenizer = MBart50Tokenizer.from_pretrained("Mudasir692/bart-urdu-summarizer")
model = MBartForConditionalGeneration.from_pretrained("Mudasir692/bart-urdu-summarizer")
Example input text (Urdu)
input_text = """
تعلیم ایک معاشرتی ترقی کا بنیادی عنصر ہے۔ حالیہ برسوں میں مختلف اداروں نے تعلیمی معیار کو بہتر بنانے اور زیادہ بچوں تک تعلیم کی رسائی ممکن بنانے کے لیے مختلف اقدامات کیے ہیں۔
ان اقدامات میں اسکولوں کی تعداد بڑھانا، اساتذہ کی تربیت میں اضافہ کرنا، اور تعلیمی مواد کی دستیابی کو یقینی بنانا شامل ہے۔ ماہرین کا خیال ہے کہ اگر یہ کوششیں مؤثر طریقے سے کی جائیں تو معاشرتی ترقی میں تیزی لائی جا سکتی ہے۔
"""
Tokenize the input text
inputs = tokenizer(input_text, return_tensors="pt")
Generate the summary
with torch.no_grad():
outputs = model.generate(**inputs)
Decode the summary and print the result
summary_text = tokenizer.decode(outputs[0], skip_special_tokens=True)
print("Summary (Urdu):", summary_text)
Training Details
Training Data
The model was fine-tuned on a custom dataset of Urdu text paired with concise summaries, focusing on headline-based examples. The dataset included a variety of topics to improve the generalization capabilities of the model.
Training Procedure
The model was fine-tuned using techniques like mixed precision to optimize training efficiency and performance.
Training Hyperparameters
Training regime: Mixed precision (fp16)
Maximum sequence length: 512
Batch size: 2
accumulation_steps = 8
Learning rate: 3e-5
Evaluation
The model's performance was evaluated using ROUGE metrics, which showed strong alignment between the generated summaries and reference summaries in the dataset.
bibtex
Copy code
@model{mudasir692_bart_urdu_summarizer,
author = {Mudasir},
title = {Bart-Urdu-Summarizer},
year = {2024},
url = {
https://huggingface.co/Mudasir692/bart-urdu-summarizer}
}
APA: Mudasir. (2024). Bart-Urdu-Summarizer. Retrieved from
https://huggingface.co/Mudasir692/bart-urdu-summarizer.
