README.md

metadata

license: apache-2.0
base_model: google/flan-t5-base
tags:
  - summarization
  - news
  - text2text-generation
  - bart-large-cnn
language:
  - en
datasets:
  - cnn_dailymail
metrics:
  - rouge
pipeline_tag: summarization

News Summarizer

This model is fine-tuned for news article summarization. It can take long news articles and generate concise, accurate summaries.

Model Details

• Base Model: facebook/bart-large-cnn
• Task: Text Summarization
• Language: English
• Training Steps: 4000
• Best ROUGE-1: 0.42
• Live version on Streamlit: https://english-news-summarizer.streamlit.app

Usage

from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
import re

# Load model
model = AutoModelForSeq2SeqLM.from_pretrained("ciorant/news-summarizer")
tokenizer = AutoTokenizer.from_pretrained("ciorant/news-summarizer")

def summarize_news(article_text, max_length=128):
    inputs = tokenizer(article_text, return_tensors="pt", truncation=True, max_length=512)
    
    outputs = model.generate(
        inputs.input_ids,
        max_length=max_length,
        num_beams=4,
        early_stopping=True,
        do_sample=False,
        length_penalty=1.0
    )
    
    summary = tokenizer.decode(outputs[0], skip_special_tokens=True)
    
    # Clean up spacing around punctuation
    summary = re.sub(r'\s+([.,!?;:])', r'\1', summary)
    summary = re.sub(r'\s+', ' ', summary)
    
    return summary.strip()

# Example usage
article = "Your news article text here..."
summary = summarize_news(article)
print(summary)

Training Data

Trained on news articles for summarization task.

Performance

• ROUGE-1: ~0.42
• ROUGE-2: ~0.21
• ROUGE-L: ~0.29

Limitations

• Optimized for English news articles
• Best performance on articles 100-800 words